leoxiaobin / deep-high-resolution-net.pytorch Goto Github PK

The project is an official implementation of our CVPR2019 paper "Deep High-Resolution Representation Learning for Human Pose Estimation"

Home Page: https://jingdongwang2017.github.io/Projects/HRNet/PoseEstimation.html

License: MIT License

Makefile 0.03% Python 31.39% Cuda 67.71% C++ 0.03% Cython 0.84%

human-pose-estimation deep-learning coco-keypoints-detection mpii-dataset mpii mscoco-keypoint deep-high-resolution-net high-resolution-net

deep-high-resolution-net.pytorch's Introduction

Deep High-Resolution Representation Learning for Human Pose Estimation (CVPR 2019)

News

[2021/04/12] Welcome to check out our recent work on bottom-up pose estimation (CVPR 2021) HRNet-DEKR!
[2020/07/05] A very nice blog from Towards Data Science introducing HRNet and HigherHRNet for human pose estimation.
[2020/03/13] A longer version is accepted by TPAMI: Deep High-Resolution Representation Learning for Visual Recognition. It includes more HRNet applications, and the codes are available: semantic segmentation, objection detection, facial landmark detection, and image classification.
[2020/02/01] We have added demo code for HRNet. Thanks Alex Simes.
Visualization code for showing the pose estimation results. Thanks Depu!
[2019/08/27] HigherHRNet is now on ArXiv, which is a bottom-up approach for human pose estimation powerd by HRNet. We will also release code and models at Higher-HRNet-Human-Pose-Estimation, stay tuned!
Our new work High-Resolution Representations for Labeling Pixels and Regions is available at HRNet. Our HRNet has been applied to a wide range of vision tasks, such as image classification, objection detection, semantic segmentation and facial landmark.

Introduction

This is an official pytorch implementation of Deep High-Resolution Representation Learning for Human Pose Estimation. In this work, we are interested in the human pose estimation problem with a focus on learning reliable high-resolution representations. Most existing methods recover high-resolution representations from low-resolution representations produced by a high-to-low resolution network. Instead, our proposed network maintains high-resolution representations through the whole process. We start from a high-resolution subnetwork as the first stage, gradually add high-to-low resolution subnetworks one by one to form more stages, and connect the mutli-resolution subnetworks in parallel. We conduct repeated multi-scale fusions such that each of the high-to-low resolution representations receives information from other parallel representations over and over, leading to rich high-resolution representations. As a result, the predicted keypoint heatmap is potentially more accurate and spatially more precise. We empirically demonstrate the effectiveness of our network through the superior pose estimation results over two benchmark datasets: the COCO keypoint detection dataset and the MPII Human Pose dataset.

Main Results

Results on MPII val

Arch	Head	Shoulder	Elbow	Wrist	Hip	Knee	Ankle	Mean	[email protected]
pose_resnet_50	96.4	95.3	89.0	83.2	88.4	84.0	79.6	88.5	34.0
pose_resnet_101	96.9	95.9	89.5	84.4	88.4	84.5	80.7	89.1	34.0
pose_resnet_152	97.0	95.9	90.0	85.0	89.2	85.3	81.3	89.6	35.0
pose_hrnet_w32	97.1	95.9	90.3	86.4	89.1	87.1	83.3	90.3	37.7

Note:

Flip test is used.
Input size is 256x256
pose_resnet_[50,101,152] is our previous work of Simple Baselines for Human Pose Estimation and Tracking

Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset

Arch	Input size	#Params	GFLOPs	AP	Ap .5	AP .75	AP (M)	AP (L)	AR	AR .5	AR .75	AR (M)	AR (L)
pose_resnet_50	256x192	34.0M	8.9	0.704	0.886	0.783	0.671	0.772	0.763	0.929	0.834	0.721	0.824
pose_resnet_50	384x288	34.0M	20.0	0.722	0.893	0.789	0.681	0.797	0.776	0.932	0.838	0.728	0.846
pose_resnet_101	256x192	53.0M	12.4	0.714	0.893	0.793	0.681	0.781	0.771	0.934	0.840	0.730	0.832
pose_resnet_101	384x288	53.0M	27.9	0.736	0.896	0.803	0.699	0.811	0.791	0.936	0.851	0.745	0.858
pose_resnet_152	256x192	68.6M	15.7	0.720	0.893	0.798	0.687	0.789	0.778	0.934	0.846	0.736	0.839
pose_resnet_152	384x288	68.6M	35.3	0.743	0.896	0.811	0.705	0.816	0.797	0.937	0.858	0.751	0.863
pose_hrnet_w32	256x192	28.5M	7.1	0.744	0.905	0.819	0.708	0.810	0.798	0.942	0.865	0.757	0.858
pose_hrnet_w32	384x288	28.5M	16.0	0.758	0.906	0.825	0.720	0.827	0.809	0.943	0.869	0.767	0.871
pose_hrnet_w48	256x192	63.6M	14.6	0.751	0.906	0.822	0.715	0.818	0.804	0.943	0.867	0.762	0.864
pose_hrnet_w48	384x288	63.6M	32.9	0.763	0.908	0.829	0.723	0.834	0.812	0.942	0.871	0.767	0.876

Note:

Flip test is used.
Person detector has person AP of 56.4 on COCO val2017 dataset.
pose_resnet_[50,101,152] is our previous work of Simple Baselines for Human Pose Estimation and Tracking.
GFLOPs is for convolution and linear layers only.

Results on COCO test-dev2017 with detector having human AP of 60.9 on COCO test-dev2017 dataset

Arch	Input size	#Params	GFLOPs	AP	Ap .5	AP .75	AP (M)	AP (L)	AR	AR .5	AR .75	AR (M)	AR (L)
pose_resnet_152	384x288	68.6M	35.3	0.737	0.919	0.828	0.713	0.800	0.790	0.952	0.856	0.748	0.849
pose_hrnet_w48	384x288	63.6M	32.9	0.755	0.925	0.833	0.719	0.815	0.805	0.957	0.874	0.763	0.863
pose_hrnet_w48*	384x288	63.6M	32.9	0.770	0.927	0.845	0.734	0.831	0.820	0.960	0.886	0.778	0.877

Note:

Flip test is used.
Person detector has person AP of 60.9 on COCO test-dev2017 dataset.
pose_resnet_152 is our previous work of Simple Baselines for Human Pose Estimation and Tracking.
GFLOPs is for convolution and linear layers only.
pose_hrnet_w48* means using additional data from AI challenger for training.

Environment

The code is developed using python 3.6 on Ubuntu 16.04. NVIDIA GPUs are needed. The code is developed and tested using 4 NVIDIA P100 GPU cards. Other platforms or GPU cards are not fully tested.

Quick start

Installation

Install pytorch >= v1.0.0 following official instruction. Note that if you use pytorch's version < v1.0.0, you should following the instruction at https://github.com/Microsoft/human-pose-estimation.pytorch to disable cudnn's implementations of BatchNorm layer. We encourage you to use higher pytorch's version(>=v1.0.0)
Clone this repo, and we'll call the directory that you cloned as ${POSE_ROOT}.
Install dependencies:
```
pip install -r requirements.txt
```
Make libs:
```
cd ${POSE_ROOT}/lib
make
```

Install COCOAPI:

# COCOAPI=/path/to/clone/cocoapi
git clone https://github.com/cocodataset/cocoapi.git $COCOAPI
cd $COCOAPI/PythonAPI
# Install into global site-packages
make install
# Alternatively, if you do not have permissions or prefer
# not to install the COCO API into global site-packages
python3 setup.py install --user

Note that instructions like # COCOAPI=/path/to/install/cocoapi indicate that you should pick a path where you'd like to have the software cloned and then set an environment variable (COCOAPI in this case) accordingly.

Init output(training model output directory) and log(tensorboard log directory) directory:

mkdir output 
mkdir log

Your directory tree should look like this:

${POSE_ROOT}
├── data
├── experiments
├── lib
├── log
├── models
├── output
├── tools 
├── README.md
└── requirements.txt

Download pretrained models from our model zoo(GoogleDrive or OneDrive)

${POSE_ROOT}
 `-- models
     `-- pytorch
         |-- imagenet
         |   |-- hrnet_w32-36af842e.pth
         |   |-- hrnet_w48-8ef0771d.pth
         |   |-- resnet50-19c8e357.pth
         |   |-- resnet101-5d3b4d8f.pth
         |   `-- resnet152-b121ed2d.pth
         |-- pose_coco
         |   |-- pose_hrnet_w32_256x192.pth
         |   |-- pose_hrnet_w32_384x288.pth
         |   |-- pose_hrnet_w48_256x192.pth
         |   |-- pose_hrnet_w48_384x288.pth
         |   |-- pose_resnet_101_256x192.pth
         |   |-- pose_resnet_101_384x288.pth
         |   |-- pose_resnet_152_256x192.pth
         |   |-- pose_resnet_152_384x288.pth
         |   |-- pose_resnet_50_256x192.pth
         |   `-- pose_resnet_50_384x288.pth
         `-- pose_mpii
             |-- pose_hrnet_w32_256x256.pth
             |-- pose_hrnet_w48_256x256.pth
             |-- pose_resnet_101_256x256.pth
             |-- pose_resnet_152_256x256.pth
             `-- pose_resnet_50_256x256.pth

Data preparation

For MPII data, please download from MPII Human Pose Dataset. The original annotation files are in matlab format. We have converted them into json format, you also need to download them from OneDrive or GoogleDrive. Extract them under {POSE_ROOT}/data, and make them look like this:

${POSE_ROOT}
|-- data
`-- |-- mpii
    `-- |-- annot
        |   |-- gt_valid.mat
        |   |-- test.json
        |   |-- train.json
        |   |-- trainval.json
        |   `-- valid.json
        `-- images
            |-- 000001163.jpg
            |-- 000003072.jpg

For COCO data, please download from COCO download, 2017 Train/Val is needed for COCO keypoints training and validation. We also provide person detection result of COCO val2017 and test-dev2017 to reproduce our multi-person pose estimation results. Please download from OneDrive or GoogleDrive. Download and extract them under {POSE_ROOT}/data, and make them look like this:

${POSE_ROOT}
|-- data
`-- |-- coco
    `-- |-- annotations
        |   |-- person_keypoints_train2017.json
        |   `-- person_keypoints_val2017.json
        |-- person_detection_results
        |   |-- COCO_val2017_detections_AP_H_56_person.json
        |   |-- COCO_test-dev2017_detections_AP_H_609_person.json
        `-- images
            |-- train2017
            |   |-- 000000000009.jpg
            |   |-- 000000000025.jpg
            |   |-- 000000000030.jpg
            |   |-- ... 
            `-- val2017
                |-- 000000000139.jpg
                |-- 000000000285.jpg
                |-- 000000000632.jpg
                |-- ...

Training and Testing

Testing on MPII dataset using model zoo's models(GoogleDrive or OneDrive)

python tools/test.py \
    --cfg experiments/mpii/hrnet/w32_256x256_adam_lr1e-3.yaml \
    TEST.MODEL_FILE models/pytorch/pose_mpii/pose_hrnet_w32_256x256.pth

Training on MPII dataset

python tools/train.py \
    --cfg experiments/mpii/hrnet/w32_256x256_adam_lr1e-3.yaml

Testing on COCO val2017 dataset using model zoo's models(GoogleDrive or OneDrive)

python tools/test.py \
    --cfg experiments/coco/hrnet/w32_256x192_adam_lr1e-3.yaml \
    TEST.MODEL_FILE models/pytorch/pose_coco/pose_hrnet_w32_256x192.pth \
    TEST.USE_GT_BBOX False

Training on COCO train2017 dataset

python tools/train.py \
    --cfg experiments/coco/hrnet/w32_256x192_adam_lr1e-3.yaml \

Visualization

Visualizing predictions on COCO val

python visualization/plot_coco.py \
    --prediction output/coco/w48_384x288_adam_lr1e-3/results/keypoints_val2017_results_0.json \
    --save-path visualization/results

Other applications

Many other dense prediction tasks, such as segmentation, face alignment and object detection, etc. have been benefited by HRNet. More information can be found at High-Resolution Networks.

Other implementation

mmpose
ModelScope (中文）
timm

Citation

If you use our code or models in your research, please cite with:

@inproceedings{sun2019deep,
  title={Deep High-Resolution Representation Learning for Human Pose Estimation},
  author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
  booktitle={CVPR},
  year={2019}
}

@inproceedings{xiao2018simple,
    author={Xiao, Bin and Wu, Haiping and Wei, Yichen},
    title={Simple Baselines for Human Pose Estimation and Tracking},
    booktitle = {European Conference on Computer Vision (ECCV)},
    year = {2018}
}

@article{WangSCJDZLMTWLX19,
  title={Deep High-Resolution Representation Learning for Visual Recognition},
  author={Jingdong Wang and Ke Sun and Tianheng Cheng and 
          Borui Jiang and Chaorui Deng and Yang Zhao and Dong Liu and Yadong Mu and 
          Mingkui Tan and Xinggang Wang and Wenyu Liu and Bin Xiao},
  journal   = {TPAMI}
  year={2019}
}

deep-high-resolution-net.pytorch's People

Contributors

Stargazers

Watchers

Forkers

xingyizhou hzhang57 liu3xing3long sunke123 hawkcoder weihuang527 chaoso pkurainbow liuguoyou column6942 ankitshah009 ml-lab peterzs aihgf meitianjinbu owenlnut tqdavid summitchen geoffzhang leihuan925 jacklongking baipdiw wuyunxiangwyx trantorrepository baiti01 acuriouscat leon0n zhe-meng jcjs hyli2020 massymeniche edmig lewiszhao jdc08161063 mathpopo kevinjunwei ieee820 dorniwang zengai tangyoubao 738654805 fd-mingjie xianpf aznikline avatarworld hdjsjyl yerikshu mellivorapku randal7 yanqi1811 gen0924 0lddriv3r zzw1123 liyang53719 sklf chencq1234 limiaopeng blankworld qingsong99 christinaliang monoloxo hcsearch zfyong rintiunse leeyangg l00228058 donproc wynmew muxinghan jinfei3459 enderych ecustboy imadcat huaifeng1993 hyfine clcarwin uptodiff guker ericzw greenteahua yuanyuangoo w-mai alice-ren xujin1184104394 liufangzheng2017 largeapp liuge1994 csjunxu lintangss jryongithub niamleeson sj-li risingdhxs jackyjsy jangocheng shaosanye collector-m boozyguo intoffset lhq0308

deep-high-resolution-net.pytorch's Issues

Low mAP obtained in testing Simple-baseline model using code in this repo.

I test the simple baseline models downloaded from your link using the code in this repo. I just get 65.3 mAP on pose_resnet50. Flip-test, Ground-truth bbox, are used, and input size is 256*192. GPU: 4 k80, batch-size:32.
And I test onresnet101, 152, it's still lower than the results in simple-baseline repo.

[Missing File (?)] 'lib/core/config.py' file to use for inference script

Hi @leoxiaobin,

Thanks again for the state-of-the-art achievement with this fine work.

Since both repos look bit similar, I am trying to follow the script here for inference on my own data:
microsoft/human-pose-estimation.pytorch#26 (comment)

This script needs 'lib/core/config.py' module, that's missing here.
I tried copy paste 'config.py' module from here and use it with the inference script above, though i get following error (among other errors):

Traceback (most recent call last):
  File "/hdd/superpharm/DHR/deep-high-resolution-net.pytorch/inference.py", line 164, in <module>
    main()
  File "/hdd/superpharm/DHR/deep-high-resolution-net.pytorch/inference.py", line 83, in main
    model.load_state_dict(torch.load(model_file))
  File "/hdd/superpharm/DHR/deep-high-resolution-net.pytorch/venv/lib/python3.5/site-packages/torch/nn/modules/module.py", line 769, in load_state_dict
    self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for PoseResNet:
	Missing key(s) in state_dict: "layer2.0.conv1.weight", "layer2.0.bn1.running_mean", "layer2.0.bn1.weight", "layer2.0.bn1.running_var", "layer2.0.bn1.bias", "layer2.0.conv2.weight", "layer2.0.bn2.running_mean", "layer2.0.bn2.weight", "layer2.0.bn2.running_var", "layer2.0.bn2.bias", "layer2.0.conv3.weight", "layer2.0.bn3.running_mean", "layer2.0.bn3.weight", "layer2.0.bn3.running_var", "layer2.0.bn3.bias", "layer2.0.downsample.0.weight", "layer2.0.downsample.1.running_mean", "layer2.0.downsample.1.weight", "layer2.0.downsample.1.running_var", "layer2.0.downsample.1.bias", "layer2.1.conv1.weight", "layer2.1.bn1.running_mean", "layer2.1.bn1.weight", "layer2.1.bn1.running_var", "layer2.1.bn1.bias", "layer2.1.conv2.weight", "layer2.1.bn2.running_mean", "layer2.1.bn2.weight", "layer2.1.bn2.running_var", "layer2.1.bn2.bias", "layer2.1.conv3.weight", "layer2.1.bn3.running_mean", "layer2.1.bn3.weight", "layer2.1.bn3.running_var", "layer2.1.bn3.bias", "layer2.2.conv1.weight", "layer2.2.bn1.running_mean", "layer2.2.bn1.weight", "layer2.2.bn1.running_var", "layer2.2.bn1.bias", "layer2.2.conv2.weight", "layer2.2.bn2.running_mean", "layer2.2.bn2.weight", "layer2.2.bn2.running_var", "layer2.2.bn2.bias", "layer2.2.conv3.weight", "layer2.2.bn3.running_mean", "layer2.2.bn3.weight", "layer2.2.bn3.running_var", "layer2.2.bn3.bias", "layer2.3.conv1.weight", "layer2.3.bn1.running_mean", "layer2.3.bn1.weight", "layer2.3.bn1.running_var", "layer2.3.bn1.bias", "layer2.3.conv2.weight", "layer2.3.bn2.running_mean", "layer2.3.bn2.weight", "layer2.3.bn2.running_var", "layer2.3.bn2.bias", "layer2.3.conv3.weight", "layer2.3.bn3.running_mean", "layer2.3.bn3.weight", "layer2.3.bn3.running_var", "layer2.3.bn3.bias", "layer3.0.conv1.weight", "layer3.0.bn1.running_mean", "layer3.0.bn1.weight", "layer3.0.bn1.running_var", "layer3.0.bn1.bias", "layer3.0.conv2.weight", "layer3.0.bn2.running_mean", "layer3.0.bn2.weight", "layer3.0.bn2.running_var", "layer3.0.bn2.bias", "layer3.0.conv3.weight", "layer3.0.bn3.running_mean", "layer3.0.bn3.weight", "layer3.0.bn3.running_var", "layer3.0.bn3.bias", "layer3.0.downsample.0.weight", "layer3.0.downsample.1.running_mean", "layer3.0.downsample.1.weight", "layer3.0.downsample.1.running_var", "layer3.0.downsample.1.bias", "layer3.1.conv1.weight", "layer3.1.bn1.running_mean", "layer3.1.bn1.weight", "layer3.1.bn1.running_var", "layer3.1.bn1.bias", "layer3.1.conv2.weight", "layer3.1.bn2.running_mean", "layer3.1.bn2.weight", "layer3.1.bn2.running_var", "layer3.1.bn2.bias", "layer3.1.conv3.weight", "layer3.1.bn3.running_mean", "layer3.1.bn3.weight", "layer3.1.bn3.running_var", "layer3.1.bn3.bias", "layer3.2.conv1.weight", "layer3.2.bn1.running_mean", "layer3.2.bn1.weight", "layer3.2.bn1.running_var", "layer3.2.bn1.bias", "layer3.2.conv2.weight", "layer3.2.bn2.running_mean", "layer3.2.bn2.weight", "layer3.2.bn2.running_var", "layer3.2.bn2.bias", "layer3.2.conv3.weight", "layer3.2.bn3.running_mean", "layer3.2.bn3.weight", "layer3.2.bn3.running_var", "layer3.2.bn3.bias", "layer3.3.conv1.weight", "layer3.3.bn1.running_mean", "layer3.3.bn1.weight", "layer3.3.bn1.running_var", "layer3.3.bn1.bias", "layer3.3.conv2.weight", "layer3.3.bn2.running_mean", "layer3.3.bn2.weight", "layer3.3.bn2.running_var", "layer3.3.bn2.bias", "layer3.3.conv3.weight", "layer3.3.bn3.running_mean", "layer3.3.bn3.weight", "layer3.3.bn3.running_var", "layer3.3.bn3.bias", "layer3.4.conv1.weight", "layer3.4.bn1.running_mean", "layer3.4.bn1.weight", "layer3.4.bn1.running_var", "layer3.4.bn1.bias", "layer3.4.conv2.weight", "layer3.4.bn2.running_mean", "layer3.4.bn2.weight", "layer3.4.bn2.running_var", "layer3.4.bn2.bias", "layer3.4.conv3.weight", "layer3.4.bn3.running_mean", "layer3.4.bn3.weight", "layer3.4.bn3.running_var", "layer3.4.bn3.bias", "layer3.5.conv1.weight", "layer3.5.bn1.running_mean", "layer3.5.bn1.weight", "layer3.5.bn1.running_var", "layer3.5.bn1.bias", "layer3.5.conv2.weight", "layer3.5.bn2.running_mean", "layer3.5.bn2.weight", "layer3.5.bn2.running_var", "layer3.5.bn2.bias", "layer3.5.conv3.weight", "layer3.5.bn3.running_mean", "layer3.5.bn3.weight", "layer3.5.bn3.running_var", "layer3.5.bn3.bias", "layer4.0.conv1.weight", "layer4.0.bn1.running_mean", "layer4.0.bn1.weight", "layer4.0.bn1.running_var", "layer4.0.bn1.bias", "layer4.0.conv2.weight", "layer4.0.bn2.running_mean", "layer4.0.bn2.weight", "layer4.0.bn2.running_var", "layer4.0.bn2.bias", "layer4.0.conv3.weight", "layer4.0.bn3.running_mean", "layer4.0.bn3.weight", "layer4.0.bn3.running_var", "layer4.0.bn3.bias", "layer4.0.downsample.0.weight", "layer4.0.downsample.1.running_mean", "layer4.0.downsample.1.weight", "layer4.0.downsample.1.running_var", "layer4.0.downsample.1.bias", "layer4.1.conv1.weight", "layer4.1.bn1.running_mean", "layer4.1.bn1.weight", "layer4.1.bn1.running_var", "layer4.1.bn1.bias", "layer4.1.conv2.weight", "layer4.1.bn2.running_mean", "layer4.1.bn2.weight", "layer4.1.bn2.running_var", "layer4.1.bn2.bias", "layer4.1.conv3.weight", "layer4.1.bn3.running_mean", "layer4.1.bn3.weight", "layer4.1.bn3.running_var", "layer4.1.bn3.bias", "layer4.2.conv1.weight", "layer4.2.bn1.running_mean", "layer4.2.bn1.weight", "layer4.2.bn1.running_var", "layer4.2.bn1.bias", "layer4.2.conv2.weight", "layer4.2.bn2.running_mean", "layer4.2.bn2.weight", "layer4.2.bn2.running_var", "layer4.2.bn2.bias", "layer4.2.conv3.weight", "layer4.2.bn3.running_mean", "layer4.2.bn3.weight", "layer4.2.bn3.running_var", "layer4.2.bn3.bias", "deconv_layers.0.weight", "deconv_layers.1.running_mean", "deconv_layers.1.weight", "deconv_layers.1.running_var", "deconv_layers.1.bias", "deconv_layers.3.weight", "deconv_layers.4.running_mean", "deconv_layers.4.weight", "deconv_layers.4.running_var", "deconv_layers.4.bias", "deconv_layers.6.weight", "deconv_layers.7.running_mean", "deconv_layers.7.weight", "deconv_layers.7.running_var", "deconv_layers.7.bias". 
	Unexpected key(s) in state_dict: "conv2.weight", "bn2.weight", "bn2.bias", "bn2.running_mean", "bn2.running_var", "bn2.num_batches_tracked", "transition1.0.0.weight", "transition1.0.1.weight", "transition1.0.1.bias", "transition1.0.1.running_mean", "transition1.0.1.running_var", "transition1.0.1.num_batches_tracked", "transition1.1.0.0.weight", "transition1.1.0.1.weight", "transition1.1.0.1.bias", "transition1.1.0.1.running_mean", "transition1.1.0.1.running_var", "transition1.1.0.1.num_batches_tracked", "stage2.0.branches.0.0.conv1.weight", "stage2.0.branches.0.0.bn1.weight", "stage2.0.branches.0.0.bn1.bias", "stage2.0.branches.0.0.bn1.running_mean", "stage2.0.branches.0.0.bn1.running_var", "stage2.0.branches.0.0.bn1.num_batches_tracked", "stage2.0.branches.0.0.conv2.weight", "stage2.0.branches.0.0.bn2.weight", "stage2.0.branches.0.0.bn2.bias", "stage2.0.branches.0.0.bn2.running_mean", "stage2.0.branches.0.0.bn2.running_var", "stage2.0.branches.0.0.bn2.num_batches_tracked", "stage2.0.branches.0.1.conv1.weight", "stage2.0.branches.0.1.bn1.weight", "stage2.0.branches.0.1.bn1.bias", "stage2.0.branches.0.1.bn1.running_mean", "stage2.0.branches.0.1.bn1.running_var", "stage2.0.branches.0.1.bn1.num_batches_tracked", "stage2.0.branches.0.1.conv2.weight", "stage2.0.branches.0.1.bn2.weight", "stage2.0.branches.0.1.bn2.bias", "stage2.0.branches.0.1.bn2.running_mean", "stage2.0.branches.0.1.bn2.running_var", "stage2.0.branches.0.1.bn2.num_batches_tracked", "stage2.0.branches.0.2.conv1.weight", "stage2.0.branches.0.2.bn1.weight", "stage2.0.branches.0.2.bn1.bias", "stage2.0.branches.0.2.bn1.running_mean", "stage2.0.branches.0.2.bn1.running_var", "stage2.0.branches.0.2.bn1.num_batches_tracked", "stage2.0.branches.0.2.conv2.weight", "stage2.0.branches.0.2.bn2.weight", "stage2.0.branches.0.2.bn2.bias", "stage2.0.branches.0.2.bn2.running_mean", "stage2.0.branches.0.2.bn2.running_var", "stage2.0.branches.0.2.bn2.num_batches_tracked", "stage2.0.branches.0.3.conv1.weight", "stage2.0.branches.0.3.bn1.weight", "stage2.0.branches.0.3.bn1.bias", "stage2.0.branches.0.3.bn1.running_mean", "stage2.0.branches.0.3.bn1.running_var", "stage2.0.branches.0.3.bn1.num_batches_tracked", "stage2.0.branches.0.3.conv2.weight", "stage2.0.branches.0.3.bn2.weight", "stage2.0.branches.0.3.bn2.bias", "stage2.0.branches.0.3.bn2.running_mean", "stage2.0.branches.0.3.bn2.running_var", "stage2.0.branches.0.3.bn2.num_batches_tracked", "stage2.0.branches.1.0.conv1.weight", "stage2.0.branches.1.0.bn1.weight", "stage2.0.branches.1.0.bn1.bias", "stage2.0.branches.1.0.bn1.running_mean", "stage2.0.branches.1.0.bn1.running_var", "stage2.0.branches.1.0.bn1.num_batches_tracked", "stage2.0.branches.1.0.conv2.weight", "stage2.0.branches.1.0.bn2.weight", "stage2.0.branches.1.0.bn2.bias", "stage2.0.branches.1.0.bn2.running_mean", "stage2.0.branches.1.0.bn2.running_var", "stage2.0.branches.1.0.bn2.num_batches_tracked", "stage2.0.branches.1.1.conv1.weight", "stage2.0.branches.1.1.bn1.weight", "stage2.0.branches.1.1.bn1.bias", "stage2.0.branches.1.1.bn1.running_mean", "stage2.0.branches.1.1.bn1.running_var", "stage2.0.branches.1.1.bn1.num_batches_tracked", "stage2.0.branches.1.1.conv2.weight", "stage2.0.branches.1.1.bn2.weight", "stage2.0.branches.1.1.bn2.bias", "stage2.0.branches.1.1.bn2.running_mean", "stage2.0.branches.1.1.bn2.running_var", "stage2.0.branches.1.1.bn2.num_batches_tracked", "stage2.0.branches.1.2.conv1.weight", "stage2.0.branches.1.2.bn1.weight", "stage2.0.branches.1.2.bn1.bias", "stage2.0.branches.1.2.bn1.running_mean", "stage2.0.branches.1.2.bn1.running_var", "stage2.0.branches.1.2.bn1.num_batches_tracked", "stage2.0.branches.1.2.conv2.weight", "stage2.0.branches.1.2.bn2.weight", "stage2.0.branches.1.2.bn2.bias", "stage2.0.branches.1.2.bn2.running_mean", "stage2.0.branches.1.2.bn2.running_var", "stage2.0.branches.1.2.bn2.num_batches_tracked", "stage2.0.branches.1.3.conv1.weight", "stage2.0.branches.1.3.bn1.weight", "stage2.0.branches.1.3.bn1.bias", "stage2.0.branches.1.3.bn1.running_mean", "stage2.0.branches.1.3.bn1.running_var", "stage2.0.branches.1.3.bn1.num_batches_tracked", "stage2.0.branches.1.3.conv2.weight", "stage2.0.branches.1.3.bn2.weight", "stage2.0.branches.1.3.bn2.bias", "stage2.0.branches.1.3.bn2.running_mean", "stage2.0.branches.1.3.bn2.running_var", "stage2.0.branches.1.3.bn2.num_batches_tracked", "stage2.0.fuse_layers.0.1.0.weight", "stage2.0.fuse_layers.0.1.1.weight", "stage2.0.fuse_layers.0.1.1.bias", "stage2.0.fuse_layers.0.1.1.running_mean", "stage2.0.fuse_layers.0.1.1.running_var", "stage2.0.fuse_layers.0.1.1.num_batches_tracked", "stage2.0.fuse_layers.1.0.0.0.weight", "stage2.0.fuse_layers.1.0.0.1.weight", "stage2.0.fuse_layers.1.0.0.1.bias", "stage2.0.fuse_layers.1.0.0.1.running_mean", "stage2.0.fuse_layers.1.0.0.1.running_var", "stage2.0.fuse_layers.1.0.0.1.num_batches_tracked", "transition2.2.0.0.weight", "transition2.2.0.1.weight", "transition2.2.0.1.bias", "transition2.2.0.1.running_mean", "transition2.2.0.1.running_var", "transition2.2.0.1.num_batches_tracked", "stage3.0.branches.0.0.conv1.weight", "stage3.0.branches.0.0.bn1.weight", "stage3.0.branches.0.0.bn1.bias", "stage3.0.branches.0.0.bn1.running_mean", "stage3.0.branches.0.0.bn1.running_var", "stage3.0.branches.0.0.bn1.num_batches_tracked", "stage3.0.branches.0.0.conv2.weight", "stage3.0.branches.0.0.bn2.weight", "stage3.0.branches.0.0.bn2.bias", "stage3.0.branches.0.0.bn2.running_mean", "stage3.0.branches.0.0.bn2.running_var", "stage3.0.branches.0.0.bn2.num_batches_tracked", "stage3.0.branches.0.1.conv1.weight", "stage3.0.branches.0.1.bn1.weight", "stage3.0.branches.0.1.bn1.bias", "stage3.0.branches.0.1.bn1.running_mean", "stage3.0.branches.0.1.bn1.running_var", "stage3.0.branches.0.1.bn1.num_batches_tracked", "stage3.0.branches.0.1.conv2.weight", "stage3.0.branches.0.1.bn2.weight", "stage3.0.branches.0.1.bn2.bias", "stage3.0.branches.0.1.bn2.running_mean", "stage3.0.branches.0.1.bn2.running_var", "stage3.0.branches.0.1.bn2.num_batches_tracked", "stage3.0.branches.0.2.conv1.weight", "stage3.0.branches.0.2.bn1.weight", "stage3.0.branches.0.2.bn1.bias", "stage3.0.branches.0.2.bn1.running_mean", "stage3.0.branches.0.2.bn1.running_var", "stage3.0.branches.0.2.bn1.num_batches_tracked", "stage3.0.branches.0.2.conv2.weight", "stage3.0.branches.0.2.bn2.weight", "stage3.0.branches.0.2.bn2.bias", "stage3.0.branches.0.2.bn2.running_mean", "stage3.0.branches.0.2.bn2.running_var", "stage3.0.branches.0.2.bn2.num_batches_tracked", "stage3.0.branches.0.3.conv1.weight", "stage3.0.branches.0.3.bn1.weight", "stage3.0.branches.0.3.bn1.bias", "stage3.0.branches.0.3.bn1.running_mean", "stage3.0.branches.0.3.bn1.running_var", "stage3.0.branches.0.3.bn1.num_batches_tracked", "stage3.0.branches.0.3.conv2.weight", "stage3.0.branches.0.3.bn2.weight", "stage3.0.branches.0.3.bn2.bias", "stage3.0.branches.0.3.bn2.running_mean", "stage3.0.branches.0.3.bn2.running_var", "stage3.0.branches.0.3.bn2.num_batches_tracked", "stage3.0.branches.1.0.conv1.weight", "stage3.0.branches.1.0.bn1.weight", "stage3.0.branches.1.0.bn1.bias", "stage3.0.branches.1.0.bn1.running_mean", "stage3.0.branches.1.0.bn1.running_var", "stage3.0.branches.1.0.bn1.num_batches_tracked", "stage3.0.branches.1.0.conv2.weight", "stage3.0.branches.1.0.bn2.weight", "stage3.0.branches.1.0.bn2.bias", "stage3.0.branches.1.0.bn2.running_mean", "stage3.0.branches.1.0.bn2.running_var", "stage3.0.branches.1.0.bn2.num_batches_tracked", "stage3.0.branches.1.1.conv1.weight", "stage3.0.branches.1.1.bn1.weight", "stage3.0.branches.1.1.bn1.bias", "stage3.0.branches.1.1.bn1.running_mean", "stage3.0.branches.1.1.bn1.running_var", "stage3.0.branches.1.1.bn1.num_batches_tracked", "stage3.0.branches.1.1.conv2.weight", "stage3.0.branches.1.1.bn2.weight", "stage3.0.branches.1.1.bn2.bias", "stage3.0.branches.1.1.bn2.running_mean", "stage3.0.branches.1.1.bn2.running_var", "stage3.0.branches.1.1.bn2.num_batches_tracked", "stage3.0.branches.1.2.conv1.weight", "stage3.0.branches.1.2.bn1.weight", "stage3.0.branches.1.2.bn1.bias", "stage3.0.branches.1.2.bn1.running_mean", "stage3.0.branches.1.2.bn1.running_var", "stage3.0.branches.1.2.bn1.num_batches_tracked", "stage3.0.branches.1.2.conv2.weight", "stage3.0.branches.1.2.bn2.weight", "stage3.0.branches.1.2.bn2.bias", "stage3.0.branches.1.2.bn2.running_mean", "stage3.0.branches.1.2.bn2.running_var", "stage3.0.branches.1.2.bn2.num_batches_tracked", "stage3.0.branches.1.3.conv1.weight", "stage3.0.branches.1.3.bn1.weight", "stage3.0.branches.1.3.bn1.bias", "stage3.0.branches.1.3.bn1.running_mean", "stage3.0.branches.1.3.bn1.running_var", "stage3.0.branches.1.3.bn1.num_batches_tracked", "stage3.0.branches.1.3.conv2.weight", "stage3.0.branches.1.3.bn2.weight", "stage3.0.branches.1.3.bn2.bias", "stage3.0.branches.1.3.bn2.running_mean", "stage3.0.branches.1.3.bn2.running_var", "stage3.0.branches.1.3.bn2.num_batches_tracked", "stage3.0.branches.2.0.conv1.weight", "stage3.0.branches.2.0.bn1.weight", "stage3.0.branches.2.0.bn1.bias", "stage3.0.branches.2.0.bn1.running_mean", "stage3.0.branches.2.0.bn1.running_var", "stage3.0.branches.2.0.bn1.num_batches_tracked", "stage3.0.branches.2.0.conv2.weight", "stage3.0.branches.2.0.bn2.weight", "stage3.0.branches.2.0.bn2.bias", "stage3.0.branches.2.0.bn2.running_mean", "stage3.0.branches.2.0.bn2.running_var", "stage3.0.branches.2.0.bn2.num_batches_tracked", "stage3.0.branches.2.1.conv1.weight", "stage3.0.branches.2.1.bn1.weight", "stage3.0.branches.2.1.bn1.bias", "stage3.0.branches.2.1.bn1.running_mean", "stage3.0.branches.2.1.bn1.running_var", "stage3.0.branches.2.1.bn1.num_batches_tracked", "stage3.0.branches.2.1.conv2.weight", "stage3.0.branches.2.1.bn2.weight", "stage3.0.branches.2.1.bn2.bias", "stage3.0.branches.2.1.bn2.running_mean", "stage3.0.branches.2.1.bn2.running_var", "stage3.0.branches.2.1.bn2.num_batches_tracked", "stage3.0.branches.2.2.conv1.weight", "stage3.0.branches.2.2.bn1.weight", "stage3.0.branches.2.2.bn1.bias", "stage3.0.branches.2.2.bn1.running_mean", "stage3.0.branches.2.2.bn1.running_var", "stage3.0.branches.2.2.bn1.num_batches_tracked", "stage3.0.branches.2.2.conv2.weight", "stage3.0.branches.2.2.bn2.weight", "stage3.0.branches.2.2.bn2.bias", "stage3.0.branches.2.2.bn2.running_mean", "stage3.0.branches.2.2.bn2.running_var", "stage3.0.branches.2.2.bn2.num_batches_tracked", "stage3.0.branches.2.3.conv1.weight", "stage3.0.branches.2.3.bn1.weight", "stage3.0.branches.2.3.bn1.bias", "stage3.0.branches.2.3.bn1.running_mean", "stage3.0.branches.2.3.bn1.running_var", "stage3.0.branches.2.3.bn1.num_batches_tracked", "stage3.0.branches.2.3.conv2.weight", "stage3.0.branches.2.3.bn2.weight", "stage3.0.branches.2.3.bn2.bias", "stage3.0.branches.2.3.bn2.running_mean", "stage3.0.branches.2.3.bn2.running_var", "stage3.0.branches.2.3.bn2.num_batches_tracked", "stage3.0.fuse_layers.0.1.0.weight", "stage3.0.fuse_layers.0.1.1.weight", "stage3.0.fuse_layers.0.1.1.bias", "stage3.0.fuse_layers.0.1.1.running_mean", "stage3.0.fuse_layers.0.1.1.running_var", "stage3.0.fuse_layers.0.1.1.num_batches_tracked", "stage3.0.fuse_layers.0.2.0.weight", "stage3.0.fuse_layers.0.2.1.weight", "stage3.0.fuse_layers.0.2.1.bias", "stage3.0.fuse_layers.0.2.1.running_mean", "stage3.0.fuse_layers.0.2.1.running_var", "stage3.0.fuse_layers.0.2.1.num_batches_tracked", "stage3.0.fuse_layers.1.0.0.0.weight", "stage3.0.fuse_layers.1.0.0.1.weight", "stage3.0.fuse_layers.1.0.0.1.bias", "stage3.0.fuse_layers.1.0.0.1.running_mean", "stage3.0.fuse_layers.1.0.0.1.running_var", "stage3.0.fuse_layers.1.0.0.1.num_batches_tracked", "stage3.0.fuse_layers.1.2.0.weight", "stage3.0.fuse_layers.1.2.1.weight", "stage3.0.fuse_layers.1.2.1.bias", "stage3.0.fuse_layers.1.2.1.running_mean", "stage3.0.fuse_layers.1.2.1.running_var", "stage3.0.fuse_layers.1.2.1.num_batches_tracked", "stage3.0.fuse_layers.2.0.0.0.weight", "stage3.0.fuse_layers.2.0.0.1.weight", "stage3.0.fuse_layers.2.0.0.1.bias", "stage3.0.fuse_layers.2.0.0.1.running_mean", "stage3.0.fuse_layers.2.0.0.1.running_var", "stage3.0.fuse_layers.2.0.0.1.num_batches_tracked", "stage3.0.fuse_layers.2.0.1.0.weight", "stage3.0.fuse_layers.2.0.1.1.weight", "stage3.0.fuse_layers.2.0.1.1.bias", "stage3.0.fuse_layers.2.0.1.1.running_mean", "stage3.0.fuse_layers.2.0.1.1.running_var", "stage3.0.fuse_layers.2.0.1.1.num_batches_tracked", "stage3.0.fuse_layers.2.1.0.0.weight", "stage3.0.fuse_layers.2.1.0.1.weight", "stage3.0.fuse_layers.2.1.0.1.bias", "stage3.0.fuse_layers.2.1.0.1.running_mean", "stage3.0.fuse_layers.2.1.0.1.running_var", "stage3.0.fuse_layers.2.1.0.1.num_batches_tracked", "stage3.1.branches.0.0.conv1.weight", "stage3.1.branches.0.0.bn1.weight", "stage3.1.branches.0.0.bn1.bias", "stage3.1.branches.0.0.bn1.running_mean", "stage3.1.branches.0.0.bn1.running_var", "stage3.1.branches.0.0.bn1.num_batches_tracked", "stage3.1.branches.0.0.conv2.weight", "stage3.1.branches.0.0.bn2.weight", "stage3.1.branches.0.0.bn2.bias", "stage3.1.branches.0.0.bn2.running_mean", "stage3.1.branches.0.0.bn2.running_var", "stage3.1.branches.0.0.bn2.num_batches_tracked", "stage3.1.branches.0.1.conv1.weight", "stage3.1.branches.0.1.bn1.weight", "stage3.1.branches.0.1.bn1.bias", "stage3.1.branches.0.1.bn1.running_mean", "stage3.1.branches.0.1.bn1.running_var", "stage3.1.branches.0.1.bn1.num_batches_tracked", "stage3.1.branches.0.1.conv2.weight", "stage3.1.branches.0.1.bn2.weight", "stage3.1.branches.0.1.bn2.bias", "stage3.1.branches.0.1.bn2.running_mean", "stage3.1.branches.0.1.bn2.running_var", "stage3.1.branches.0.1.bn2.num_batches_tracked", "stage3.1.branches.0.2.conv1.weight", "stage3.1.branches.0.2.bn1.weight", "stage3.1.branches.0.2.bn1.bias", "stage3.1.branches.0.2.bn1.running_mean", "stage3.1.branches.0.2.bn1.running_var", "stage3.1.branches.0.2.bn1.num_batches_tracked", "stage3.1.branches.0.2.conv2.weight", "stage3.1.branches.0.2.bn2.weight", "stage3.1.branches.0.2.bn2.bias", "stage3.1.branches.0.2.bn2.running_mean", "stage3.1.branches.0.2.bn2.running_var", "stage3.1.branches.0.2.bn2.num_batches_tracked", "stage3.1.branches.0.3.conv1.weight", "stage3.1.branches.0.3.bn1.weight", "stage3.1.branches.0.3.bn1.bias", "stage3.1.branches.0.3.bn1.running_mean", "stage3.1.branches.0.3.bn1.running_var", "stage3.1.branches.0.3.bn1.num_batches_tracked", "stage3.1.branches.0.3.conv2.weight", "stage3.1.branches.0.3.bn2.weight", "stage3.1.branches.0.3.bn2.bias", "stage3.1.branches.0.3.bn2.running_mean", "stage3.1.branches.0.3.bn2.running_var", "stage3.1.branches.0.3.bn2.num_batches_tracked", "stage3.1.branches.1.0.conv1.weight", "stage3.1.branches.1.0.bn1.weight", "stage3.1.branches.1.0.bn1.bias", "stage3.1.branches.1.0.bn1.running_mean", "stage3.1.branches.1.0.bn1.running_var", "stage3.1.branches.1.0.bn1.num_batches_tracked", "stage3.1.branches.1.0.conv2.weight", "stage3.1.branches.1.0.bn2.weight", "stage3.1.branches.1.0.bn2.bias", "stage3.1.branches.1.0.bn2.running_mean", "stage3.1.branches.1.0.bn2.running_var", "stage3.1.branches.1.0.bn2.num_batches_tracked", "stage3.1.branches.1.1.conv1.weight", "stage3.1.branches.1.1.bn1.weight", "stage3.1.branches.1.1.bn1.bias", "stage3.1.branches.1.1.bn1.running_mean", "stage3.1.branches.1.1.bn1.running_var", "stage3.1.branches.1.1.bn1.num_batches_tracked", "stage3.1.branches.1.1.conv2.weight", "stage3.1.branches.1.1.bn2.weight", "stage3.1.branches.1.1.bn2.bias", "stage3.1.branches.1.1.bn2.running_mean", "stage3.1.branches.1.1.bn2.running_var", "stage3.1.branches.1.1.bn2.num_batches_tracked", "stage3.1.branches.1.2.conv1.weight", "stage3.1.branches.1.2.bn1.weight", "stage3.1.branches.1.2.bn1.bias", "stage3.1.branches.1.2.bn1.running_mean", "stage3.1.branches.1.2.bn1.running_var", "stage3.1.branches.1.2.bn1.num_batches_tracked", "stage3.1.branches.1.2.conv2.weight", "stage3.1.branches.1.2.bn2.weight", "stage3.1.branches.1.2.bn2.bias", "stage3.1.branches.1.2.bn2.running_mean", "stage3.1.branches.1.2.bn2.running_var", "stage3.1.branches.1.2.bn2.num_batches_tracked", "stage3.1.branches.1.3.conv1.weight", "stage3.1.branches.1.3.bn1.weight", "stage3.1.branches.1.3.bn1.bias", "stage3.1.branches.1.3.bn1.running_mean", "stage3.1.branches.1.3.bn1.running_var", "stage3.1.branches.1.3.bn1.num_batches_tracked", "stage3.1.branches.1.3.conv2.weight", "stage3.1.branches.1.3.bn2.weight", "stage3.1.branches.1.3.bn2.bias", "stage3.1.branches.1.3.bn2.running_mean", "stage3.1.branches.1.3.bn2.running_var", "stage3.1.branches.1.3.bn2.num_batches_tracked", "stage3.1.branches.2.0.conv1.weight", "stage3.1.branches.2.0.bn1.weight", "stage3.1.branches.2.0.bn1.bias", "stage3.1.branches.2.0.bn1.running_mean", "stage3.1.branches.2.0.bn1.running_var", "stage3.1.branches.2.0.bn1.num_batches_tracked", "stage3.1.branches.2.0.conv2.weight", "stage3.1.branches.2.0.bn2.weight", "stage3.1.branches.2.0.bn2.bias", "stage3.1.branches.2.0.bn2.running_mean", "stage3.1.branches.2.0.bn2.running_var", "stage3.1.branches.2.0.bn2.num_batches_tracked", "stage3.1.branches.2.1.conv1.weight", "stage3.1.branches.2.1.bn1.weight", "stage3.1.branches.2.1.bn1.bias", "stage3.1.branches.2.1.bn1.running_mean", "stage3.1.branches.2.1.bn1.running_var", "stage3.1.branches.2.1.bn1.num_batches_tracked", "stage3.1.branches.2.1.conv2.weight", "stage3.1.branches.2.1.bn2.weight", "stage3.1.branches.2.1.bn2.bias", "stage3.1.branches.2.1.bn2.running_mean", "stage3.1.branches.2.1.bn2.running_var", "stage3.1.branches.2.1.bn2.num_batches_tracked", "stage3.1.branches.2.2.conv1.weight", "stage3.1.branches.2.2.bn1.weight", "stage3.1.branches.2.2.bn1.bias", "stage3.1.branches.2.2.bn1.running_mean", "stage3.1.branches.2.2.bn1.running_var", "stage3.1.branches.2.2.bn1.num_batches_tracked", "stage3.1.branches.2.2.conv2.weight", "stage3.1.branches.2.2.bn2.weight", "stage3.1.branches.2.2.bn2.bias", "stage3.1.branches.2.2.bn2.running_mean", "stage3.1.branches.2.2.bn2.running_var", "stage3.1.branches.2.2.bn2.num_batches_tracked", "stage3.1.branches.2.3.conv1.weight", "stage3.1.branches.2.3.bn1.weight", "stage3.1.branches.2.3.bn1.bias", "stage3.1.branches.2.3.bn1.running_mean", "stage3.1.branches.2.3.bn1.running_var", "stage3.1.branches.2.3.bn1.num_batches_tracked", "stage3.1.branches.2.3.conv2.weight", "stage3.1.branches.2.3.bn2.weight", "stage3.1.branches.2.3.bn2.bias", "stage3.1.branches.2.3.bn2.running_mean", "stage3.1.branches.2.3.bn2.running_var", "stage3.1.branches.2.3.bn2.num_batches_tracked", "stage3.1.fuse_layers.0.1.0.weight", "stage3.1.fuse_layers.0.1.1.weight", "stage3.1.fuse_layers.0.1.1.bias", "stage3.1.fuse_layers.0.1.1.running_mean", "stage3.1.fuse_layers.0.1.1.running_var", "stage3.1.fuse_layers.0.1.1.num_batches_tracked", "stage3.1.fuse_layers.0.2.0.weight", "stage3.1.fuse_layers.0.2.1.weight", "stage3.1.fuse_layers.0.2.1.bias", "stage3.1.fuse_layers.0.2.1.running_mean", "stage3.1.fuse_layers.0.2.1.running_var", "stage3.1.fuse_layers.0.2.1.num_batches_tracked", "stage3.1.fuse_layers.1.0.0.0.weight", "stage3.1.fuse_layers.1.0.0.1.weight", "stage3.1.fuse_layers.1.0.0.1.bias", "stage3.1.fuse_layers.1.0.0.1.running_mean", "stage3.1.fuse_layers.1.0.0.1.running_var", "stage3.1.fuse_layers.1.0.0.1.num_batches_tracked", "stage3.1.fuse_layers.1.2.0.weight", "stage3.1.fuse_layers.1.2.1.weight", "stage3.1.fuse_layers.1.2.1.bias", "stage3.1.fuse_layers.1.2.1.running_mean", "stage3.1.fuse_layers.1.2.1.running_var", "stage3.1.fuse_layers.1.2.1.num_batches_tracked", "stage3.1.fuse_layers.2.0.0.0.weight", "stage3.1.fuse_layers.2.0.0.1.weight", "stage3.1.fuse_layers.2.0.0.1.bias", "stage3.1.fuse_layers.2.0.0.1.running_mean", "stage3.1.fuse_layers.2.0.0.1.running_var", "stage3.1.fuse_layers.2.0.0.1.num_batches_tracked", "stage3.1.fuse_layers.2.0.1.0.weight", "stage3.1.fuse_layers.2.0.1.1.weight", "stage3.1.fuse_layers.2.0.1.1.bias", "stage3.1.fuse_layers.2.0.1.1.running_mean", "stage3.1.fuse_layers.2.0.1.1.running_var", "stage3.1.fuse_layers.2.0.1.1.num_batches_tracked", "stage3.1.fuse_layers.2.1.0.0.weight", "stage3.1.fuse_layers.2.1.0.1.weight", "stage3.1.fuse_layers.2.1.0.1.bias", "stage3.1.fuse_layers.2.1.0.1.running_mean", "stage3.1.fuse_layers.2.1.0.1.running_var", "stage3.1.fuse_layers.2.1.0.1.num_batches_tracked", "stage3.2.branches.0.0.conv1.weight", "stage3.2.branches.0.0.bn1.weight", "stage3.2.branches.0.0.bn1.bias", "stage3.2.branches.0.0.bn1.running_mean", "stage3.2.branches.0.0.bn1.running_var", "stage3.2.branches.0.0.bn1.num_batches_tracked", "stage3.2.branches.0.0.conv2.weight", "stage3.2.branches.0.0.bn2.weight", "stage3.2.branches.0.0.bn2.bias", "stage3.2.branches.0.0.bn2.running_mean", "stage3.2.branches.0.0.bn2.running_var", "stage3.2.branches.0.0.bn2.num_batches_tracked", "stage3.2.branches.0.1.conv1.weight", "stage3.2.branches.0.1.bn1.weight", "stage3.2.branches.0.1.bn1.bias", "stage3.2.branches.0.1.bn1.running_mean", "stage3.2.branches.0.1.bn1.running_var", "stage3.2.branches.0.1.bn1.num_batches_tracked", "stage3.2.branches.0.1.conv2.weight", "stage3.2.branches.0.1.bn2.weight", "stage3.2.branches.0.1.bn2.bias", "stage3.2.branches.0.1.bn2.running_mean", "stage3.2.branches.0.1.bn2.running_var", "stage3.2.branches.0.1.bn2.num_batches_tracked", "stage3.2.branches.0.2.conv1.weight", "stage3.2.branches.0.2.bn1.weight", "stage3.2.branches.0.2.bn1.bias", "stage3.2.branches.0.2.bn1.running_mean", "stage3.2.branches.0.2.bn1.running_var", "stage3.2.branches.0.2.bn1.num_batches_tracked", "stage3.2.branches.0.2.conv2.weight", "stage3.2.branches.0.2.bn2.weight", "stage3.2.branches.0.2.bn2.bias", "stage3.2.branches.0.2.bn2.running_mean", "stage3.2.branches.0.2.bn2.running_var", "stage3.2.branches.0.2.bn2.num_batches_tracked", "stage3.2.branches.0.3.conv1.weight", "stage3.2.branches.0.3.bn1.weight", "stage3.2.branches.0.3.bn1.bias", "stage3.2.branches.0.3.bn1.running_mean", "stage3.2.branches.0.3.bn1.running_var", "stage3.2.branches.0.3.bn1.num_batches_tracked", "stage3.2.branches.0.3.conv2.weight", "stage3.2.branches.0.3.bn2.weight", "stage3.2.branches.0.3.bn2.bias", "stage3.2.branches.0.3.bn2.running_mean", "stage3.2.branches.0.3.bn2.running_var", "stage3.2.branches.0.3.bn2.num_batches_tracked", "stage3.2.branches.1.0.conv1.weight", "stage3.2.branches.1.0.bn1.weight", "stage3.2.branches.1.0.bn1.bias", "stage3.2.branches.1.0.bn1.running_mean", "stage3.2.branches.1.0.bn1.running_var", "stage3.2.branches.1.0.bn1.num_batches_tracked", "stage3.2.branches.1.0.conv2.weight", "stage3.2.branches.1.0.bn2.weight", "stage3.2.branches.1.0.bn2.bias", "stage3.2.branches.1.0.bn2.running_mean", "stage3.2.branches.1.0.bn2.running_var", "stage3.2.branches.1.0.bn2.num_batches_tracked", "stage3.2.branches.1.1.conv1.weight", "stage3.2.branches.1.1.bn1.weight", "stage3.2.branches.1.1.bn1.bias", "stage3.2.branches.1.1.bn1.running_mean", "stage3.2.branches.1.1.bn1.running_var", "stage3.2.branches.1.1.bn1.num_batches_tracked", "stage3.2.branches.1.1.conv2.weight", "stage3.2.branches.1.1.bn2.weight", "stage3.2.branches.1.1.bn2.bias", "stage3.2.branches.1.1.bn2.running_mean", "stage3.2.branches.1.1.bn2.running_var", "stage3.2.branches.1.1.bn2.num_batches_tracked", "stage3.2.branches.1.2.conv1.weight", "stage3.2.branches.1.2.bn1.weight", "stage3.2.branches.1.2.bn1.bias", "stage3.2.branches.1.2.bn1.running_mean", "stage3.2.branches.1.2.bn1.running_var", "stage3.2.branches.1.2.bn1.num_batches_tracked", "stage3.2.branches.1.2.conv2.weight", "stage3.2.branches.1.2.bn2.weight", "stage3.2.branches.1.2.bn2.bias", "stage3.2.branches.1.2.bn2.running_mean", "stage3.2.branches.1.2.bn2.running_var", "stage3.2.branches.1.2.bn2.num_batches_tracked", "stage3.2.branches.1.3.conv1.weight", "stage3.2.branches.1.3.bn1.weight", "stage3.2.branches.1.3.bn1.bias", "stage3.2.branches.1.3.bn1.running_mean", "stage3.2.branches.1.3.bn1.running_var", "stage3.2.branches.1.3.bn1.num_batches_tracked", "stage3.2.branches.1.3.conv2.weight", "stage3.2.branches.1.3.bn2.weight", "stage3.2.branches.1.3.bn2.bias", "stage3.2.branches.1.3.bn2.running_mean", "stage3.2.branches.1.3.bn2.running_var", "stage3.2.branches.1.3.bn2.num_batches_tracked", "stage3.2.branches.2.0.conv1.weight", "stage3.2.branches.2.0.bn1.weight", "stage3.2.branches.2.0.bn1.bias", "stage3.2.branches.2.0.bn1.running_mean", "stage3.2.branches.2.0.bn1.running_var", "stage3.2.branches.2.0.bn1.num_batches_tracked", "stage3.2.branches.2.0.conv2.weight", "stage3.2.branches.2.0.bn2.weight", "stage3.2.branches.2.0.bn2.bias", "stage3.2.branches.2.0.bn2.running_mean", "stage3.2.branches.2.0.bn2.running_var", "stage3.2.branches.2.0.bn2.num_batches_tracked", "stage3.2.branches.2.1.conv1.weight", "stage3.2.branches.2.1.bn1.weight", "stage3.2.branches.2.1.bn1.bias", "stage3.2.branches.2.1.bn1.running_mean", "stage3.2.branches.2.1.bn1.running_var", "stage3.2.branches.2.1.bn1.num_batches_tracked", "stage3.2.branches.2.1.conv2.weight", "stage3.2.branches.2.1.bn2.weight", "stage3.2.branches.2.1.bn2.bias", "stage3.2.branches.2.1.bn2.running_mean", "stage3.2.branches.2.1.bn2.running_var", "stage3.2.branches.2.1.bn2.num_batches_tracked", "stage3.2.branches.2.2.conv1.weight", "stage3.2.branches.2.2.bn1.weight", "stage3.2.branches.2.2.bn1.bias", "stage3.2.branches.2.2.bn1.running_mean", "stage3.2.branches.2.2.bn1.running_var", "stage3.2.branches.2.2.bn1.num_batches_tracked", "stage3.2.branches.2.2.conv2.weight", "stage3.2.branches.2.2.bn2.weight", "stage3.2.branches.2.2.bn2.bias", "stage3.2.branches.2.2.bn2.running_mean", "stage3.2.branches.2.2.bn2.running_var", "stage3.2.branches.2.2.bn2.num_batches_tracked", "stage3.2.branches.2.3.conv1.weight", "stage3.2.branches.2.3.bn1.weight", "stage3.2.branches.2.3.bn1.bias", "stage3.2.branches.2.3.bn1.running_mean", "stage3.2.branches.2.3.bn1.running_var", "stage3.2.branches.2.3.bn1.num_batches_tracked", "stage3.2.branches.2.3.conv2.weight", "stage3.2.branches.2.3.bn2.weight", "stage3.2.branches.2.3.bn2.bias", "stage3.2.branches.2.3.bn2.running_mean", "stage3.2.branches.2.3.bn2.running_var", "stage3.2.branches.2.3.bn2.num_batches_tracked", "stage3.2.fuse_layers.0.1.0.weight", "stage3.2.fuse_layers.0.1.1.weight", "stage3.2.fuse_layers.0.1.1.bias", "stage3.2.fuse_layers.0.1.1.running_mean", "stage3.2.fuse_layers.0.1.1.running_var", "stage3.2.fuse_layers.0.1.1.num_batches_tracked", "stage3.2.fuse_layers.0.2.0.weight", "stage3.2.fuse_layers.0.2.1.weight", "stage3.2.fuse_layers.0.2.1.bias", "stage3.2.fuse_layers.0.2.1.running_mean", "stage3.2.fuse_layers.0.2.1.running_var", "stage3.2.fuse_layers.0.2.1.num_batches_tracked", "stage3.2.fuse_layers.1.0.0.0.weight", "stage3.2.fuse_layers.1.0.0.1.weight", "stage3.2.fuse_layers.1.0.0.1.bias", "stage3.2.fuse_layers.1.0.0.1.running_mean", "stage3.2.fuse_layers.1.0.0.1.running_var", "stage3.2.fuse_layers.1.0.0.1.num_batches_tracked", "stage3.2.fuse_layers.1.2.0.weight", "stage3.2.fuse_layers.1.2.1.weight", "stage3.2.fuse_layers.1.2.1.bias", "stage3.2.fuse_layers.1.2.1.running_mean", "stage3.2.fuse_layers.1.2.1.running_var", "stage3.2.fuse_layers.1.2.1.num_batches_tracked", "stage3.2.fuse_layers.2.0.0.0.weight", "stage3.2.fuse_layers.2.0.0.1.weight", "stage3.2.fuse_layers.2.0.0.1.bias", "stage3.2.fuse_layers.2.0.0.1.running_mean", "stage3.2.fuse_layers.2.0.0.1.running_var", "stage3.2.fuse_layers.2.0.0.1.num_batches_tracked", "stage3.2.fuse_layers.2.0.1.0.weight", "stage3.2.fuse_layers.2.0.1.1.weight", "stage3.2.fuse_layers.2.0.1.1.bias", "stage3.2.fuse_layers.2.0.1.1.running_mean", "stage3.2.fuse_layers.2.0.1.1.running_var", "stage3.2.fuse_layers.2.0.1.1.num_batches_tracked", "stage3.2.fuse_layers.2.1.0.0.weight", "stage3.2.fuse_layers.2.1.0.1.weight", "stage3.2.fuse_layers.2.1.0.1.bias", "stage3.2.fuse_layers.2.1.0.1.running_mean", "stage3.2.fuse_layers.2.1.0.1.running_var", "stage3.2.fuse_layers.2.1.0.1.num_batches_tracked", "stage3.3.branches.0.0.conv1.weight", "stage3.3.branches.0.0.bn1.weight", "stage3.3.branches.0.0.bn1.bias", "stage3.3.branches.0.0.bn1.running_mean", "stage3.3.branches.0.0.bn1.running_var", "stage3.3.branches.0.0.bn1.num_batches_tracked", "stage3.3.branches.0.0.conv2.weight", "stage3.3.branches.0.0.bn2.weight", "stage3.3.branches.0.0.bn2.bias", "stage3.3.branches.0.0.bn2.running_mean", "stage3.3.branches.0.0.bn2.running_var", "stage3.3.branches.0.0.bn2.num_batches_tracked", "stage3.3.branches.0.1.conv1.weight", "stage3.3.branches.0.1.bn1.weight", "stage3.3.branches.0.1.bn1.bias", "stage3.3.branches.0.1.bn1.running_mean", "stage3.3.branches.0.1.bn1.running_var", "stage3.3.branches.0.1.bn1.num_batches_tracked", "stage3.3.branches.0.1.conv2.weight", "stage3.3.branches.0.1.bn2.weight", "stage3.3.branches.0.1.bn2.bias", "stage3.3.branches.0.1.bn2.running_mean", "stage3.3.branches.0.1.bn2.running_var", "stage3.3.branches.0.1.bn2.num_batches_tracked", "stage3.3.branches.0.2.conv1.weight", "stage3.3.branches.0.2.bn1.weight", "stage3.3.branches.0.2.bn1.bias", "stage3.3.branches.0.2.bn1.running_mean", "stage3.3.branches.0.2.bn1.running_var", "stage3.3.branches.0.2.bn1.num_batches_tracked", "stage3.3.branches.0.2.conv2.weight", "stage3.3.branches.0.2.bn2.weight", "stage3.3.branches.0.2.bn2.bias", "stage3.3.branches.0.2.bn2.running_mean", "stage3.3.branches.0.2.bn2.running_var", "stage3.3.branches.0.2.bn2.num_batches_tracked", "stage3.3.branches.0.3.conv1.weight", "stage3.3.branches.0.3.bn1.weight", "stage3.3.branches.0.3.bn1.bias", "stage3.3.branches.0.3.bn1.running_mean", "stage3.3.branches.0.3.bn1.running_var", "stage3.3.branches.0.3.bn1.num_batches_tracked", "stage3.3.branches.0.3.conv2.weight", "stage3.3.branches.0.3.bn2.weight", "stage3.3.branches.0.3.bn2.bias", "stage3.3.branches.0.3.bn2.running_mean", "stage3.3.branches.0.3.bn2.running_var", "stage3.3.branches.0.3.bn2.num_batches_tracked", "stage3.3.branches.1.0.conv1.weight", "stage3.3.branches.1.0.bn1.weight", "stage3.3.branches.1.0.bn1.bias", "stage3.3.branches.1.0.bn1.running_mean", "stage3.3.branches.1.0.bn1.running_var", "stage3.3.branches.1.0.bn1.num_batches_tracked", "stage3.3.branches.1.0.conv2.weight", "stage3.3.branches.1.0.bn2.weight", "stage3.3.branches.1.0.bn2.bias", "stage3.3.branches.1.0.bn2.running_mean", "stage3.3.branches.1.0.bn2.running_var", "stage3.3.branches.1.0.bn2.num_batches_tracked", "stage3.3.branches.1.1.conv1.weight", "stage3.3.branches.1.1.bn1.weight", "stage3.3.branches.1.1.bn1.bias", "stage3.3.branches.1.1.bn1.running_mean", "stage3.3.branches.1.1.bn1.running_var", "stage3.3.branches.1.1.bn1.num_batches_tracked", "stage3.3.branches.1.1.conv2.weight", "stage3.3.branches.1.1.bn2.weight", "stage3.3.branches.1.1.bn2.bias", "stage3.3.branches.1.1.bn2.running_mean", "stage3.3.branches.1.1.bn2.running_var", "stage3.3.branches.1.1.bn2.num_batches_tracked", "stage3.3.branches.1.2.conv1.weight", "stage3.3.branches.1.2.bn1.weight", "stage3.3.branches.1.2.bn1.bias", "stage3.3.branches.1.2.bn1.running_mean", "stage3.3.branches.1.2.bn1.running_var", "stage3.3.branches.1.2.bn1.num_batches_tracked", "stage3.3.branches.1.2.conv2.weight", "stage3.3.branches.1.2.bn2.weight", "stage3.3.branches.1.2.bn2.bias", "stage3.3.branches.1.2.bn2.running_mean", "stage3.3.branches.1.2.bn2.running_var", "stage3.3.branches.1.2.bn2.num_batches_tracked", "stage3.3.branches.1.3.conv1.weight", "stage3.3.branches.1.3.bn1.weight", "stage3.3.branches.1.3.bn1.bias", "stage3.3.branches.1.3.bn1.running_mean", "stage3.3.branches.1.3.bn1.running_var", "stage3.3.branches.1.3.bn1.num_batches_tracked", "stage3.3.branches.1.3.conv2.weight", "stage3.3.branches.1.3.bn2.weight", "stage3.3.branches.1.3.bn2.bias", "stage3.3.branches.1.3.bn2.running_mean", "stage3.3.branches.1.3.bn2.running_var", "stage3.3.branches.1.3.bn2.num_batches_tracked", "stage3.3.branches.2.0.conv1.weight", "stage3.3.branches.2.0.bn1.weight", "stage3.3.branches.2.0.bn1.bias", "stage3.3.branches.2.0.bn1.running_mean", "stage3.3.branches.2.0.bn1.running_var", "stage3.3.branches.2.0.bn1.num_batches_tracked", "stage3.3.branches.2.0.conv2.weight", "stage3.3.branches.2.0.bn2.weight", "stage3.3.branches.2.0.bn2.bias", "stage3.3.branches.2.0.bn2.running_mean", "stage3.3.branches.2.0.bn2.running_var", "stage3.3.branches.2.0.bn2.num_batches_tracked", "stage3.3.branches.2.1.conv1.weight", "stage3.3.branches.2.1.bn1.weight", "stage3.3.branches.2.1.bn1.bias", "stage3.3.branches.2.1.bn1.running_mean", "stage3.3.branches.2.1.bn1.running_var", "stage3.3.branches.2.1.bn1.num_batches_tracked", "stage3.3.branches.2.1.conv2.weight", "stage3.3.branches.2.1.bn2.weight", "stage3.3.branches.2.1.bn2.bias", "stage3.3.branches.2.1.bn2.running_mean", "stage3.3.branches.2.1.bn2.running_var", "stage3.3.branches.2.1.bn2.num_batches_tracked", "stage3.3.branches.2.2.conv1.weight", "stage3.3.branches.2.2.bn1.weight", "stage3.3.branches.2.2.bn1.bias", "stage3.3.branches.2.2.bn1.running_mean", "stage3.3.branches.2.2.bn1.running_var", "stage3.3.branches.2.2.bn1.num_batches_tracked", "stage3.3.branches.2.2.conv2.weight", "stage3.3.branches.2.2.bn2.weight", "stage3.3.branches.2.2.bn2.bias", "stage3.3.branches.2.2.bn2.running_mean", "stage3.3.branches.2.2.bn2.running_var", "stage3.3.branches.2.2.bn2.num_batches_tracked", "stage3.3.branches.2.3.conv1.weight", "stage3.3.branches.2.3.bn1.weight", "stage3.3.branches.2.3.bn1.bias", "stage3.3.branches.2.3.bn1.running_mean", "stage3.3.branches.2.3.bn1.running_var", "stage3.3.branches.2.3.bn1.num_batches_tracked", "stage3.3.branches.2.3.conv2.weight", "stage3.3.branches.2.3.bn2.weight", "stage3.3.branches.2.3.bn2.bias", "stage3.3.branches.2.3.bn2.running_mean", "stage3.3.branches.2.3.bn2.running_var", "stage3.3.branches.2.3.bn2.num_batches_tracked", "stage3.3.fuse_layers.0.1.0.weight", "stage3.3.fuse_layers.0.1.1.weight", "stage3.3.fuse_layers.0.1.1.bias", "stage3.3.fuse_layers.0.1.1.running_mean", "stage3.3.fuse_layers.0.1.1.running_var", "stage3.3.fuse_layers.0.1.1.num_batches_tracked", "stage3.3.fuse_layers.0.2.0.weight", "stage3.3.fuse_layers.0.2.1.weight", "stage3.3.fuse_layers.0.2.1.bias", "stage3.3.fuse_layers.0.2.1.running_mean", "stage3.3.fuse_layers.0.2.1.running_var", "stage3.3.fuse_layers.0.2.1.num_batches_tracked", "stage3.3.fuse_layers.1.0.0.0.weight", "stage3.3.fuse_layers.1.0.0.1.weight", "stage3.3.fuse_layers.1.0.0.1.bias", "stage3.3.fuse_layers.1.0.0.1.running_mean", "stage3.3.fuse_layers.1.0.0.1.running_var", "stage3.3.fuse_layers.1.0.0.1.num_batches_tracked", "stage3.3.fuse_layers.1.2.0.weight", "stage3.3.fuse_layers.1.2.1.weight", "stage3.3.fuse_layers.1.2.1.bias", "stage3.3.fuse_layers.1.2.1.running_mean", "stage3.3.fuse_layers.1.2.1.running_var", "stage3.3.fuse_layers.1.2.1.num_batches_tracked", "stage3.3.fuse_layers.2.0.0.0.weight", "stage3.3.fuse_layers.2.0.0.1.weight", "stage3.3.fuse_layers.2.0.0.1.bias", "stage3.3.fuse_layers.2.0.0.1.running_mean", "stage3.3.fuse_layers.2.0.0.1.running_var", "stage3.3.fuse_layers.2.0.0.1.num_batches_tracked", "stage3.3.fuse_layers.2.0.1.0.weight", "stage3.3.fuse_layers.2.0.1.1.weight", "stage3.3.fuse_layers.2.0.1.1.bias", "stage3.3.fuse_layers.2.0.1.1.running_mean", "stage3.3.fuse_layers.2.0.1.1.running_var", "stage3.3.fuse_layers.2.0.1.1.num_batches_tracked", "stage3.3.fuse_layers.2.1.0.0.weight", "stage3.3.fuse_layers.2.1.0.1.weight", "stage3.3.fuse_layers.2.1.0.1.bias", "stage3.3.fuse_layers.2.1.0.1.running_mean", "stage3.3.fuse_layers.2.1.0.1.running_var", "stage3.3.fuse_layers.2.1.0.1.num_batches_tracked", "transition3.3.0.0.weight", "transition3.3.0.1.weight", "transition3.3.0.1.bias", "transition3.3.0.1.running_mean", "transition3.3.0.1.running_var", "transition3.3.0.1.num_batches_tracked", "stage4.0.branches.0.0.conv1.weight", "stage4.0.branches.0.0.bn1.weight", "stage4.0.branches.0.0.bn1.bias", "stage4.0.branches.0.0.bn1.running_mean", "stage4.0.branches.0.0.bn1.running_var", "stage4.0.branches.0.0.bn1.num_batches_tracked", "stage4.0.branches.0.0.conv2.weight", "stage4.0.branches.0.0.bn2.weight", "stage4.0.branches.0.0.bn2.bias", "stage4.0.branches.0.0.bn2.running_mean", "stage4.0.branches.0.0.bn2.running_var", "stage4.0.branches.0.0.bn2.num_batches_tracked", "stage4.0.branches.0.1.conv1.weight", "stage4.0.branches.0.1.bn1.weight", "stage4.0.branches.0.1.bn1.bias", "stage4.0.branches.0.1.bn1.running_mean", "stage4.0.branches.0.1.bn1.running_var", "stage4.0.branches.0.1.bn1.num_batches_tracked", "stage4.0.branches.0.1.conv2.weight", "stage4.0.branches.0.1.bn2.weight", "stage4.0.branches.0.1.bn2.bias", "stage4.0.branches.0.1.bn2.running_mean", "stage4.0.branches.0.1.bn2.running_var", "stage4.0.branches.0.1.bn2.num_batches_tracked", "stage4.0.branches.0.2.conv1.weight", "stage4.0.branches.0.2.bn1.weight", "stage4.0.branches.0.2.bn1.bias", "stage4.0.branches.0.2.bn1.running_mean", "stage4.0.branches.0.2.bn1.running_var", "stage4.0.branches.0.2.bn1.num_batches_tracked", "stage4.0.branches.0.2.conv2.weight", "stage4.0.branches.0.2.bn2.weight", "stage4.0.branches.0.2.bn2.bias", "stage4.0.branches.0.2.bn2.running_mean", "stage4.0.branches.0.2.bn2.running_var", "stage4.0.branches.0.2.bn2.num_batches_tracked", "stage4.0.branches.0.3.conv1.weight", "stage4.0.branches.0.3.bn1.weight", "stage4.0.branches.0.3.bn1.bias", "stage4.0.branches.0.3.bn1.running_mean", "stage4.0.branches.0.3.bn1.running_var", "stage4.0.branches.0.3.bn1.num_batches_tracked", "stage4.0.branches.0.3.conv2.weight", "stage4.0.branches.0.3.bn2.weight", "stage4.0.branches.0.3.bn2.bias", "stage4.0.branches.0.3.bn2.running_mean", "stage4.0.branches.0.3.bn2.running_var", "stage4.0.branches.0.3.bn2.num_batches_tracked", "stage4.0.branches.1.0.conv1.weight", "stage4.0.branches.1.0.bn1.weight", "stage4.0.branches.1.0.bn1.bias", "stage4.0.branches.1.0.bn1.running_mean", "stage4.0.branches.1.0.bn1.running_var", "stage4.0.branches.1.0.bn1.num_batches_tracked", "stage4.0.branches.1.0.conv2.weight", "stage4.0.branches.1.0.bn2.weight", "stage4.0.branches.1.0.bn2.bias", "stage4.0.branches.1.0.bn2.running_mean", "stage4.0.branches.1.0.bn2.running_var", "stage4.0.branches.1.0.bn2.num_batches_tracked", "stage4.0.branches.1.1.conv1.weight", "stage4.0.branches.1.1.bn1.weight", "stage4.0.branches.1.1.bn1.bias", "stage4.0.branches.1.1.bn1.running_mean", "stage4.0.branches.1.1.bn1.running_var", "stage4.0.branches.1.1.bn1.num_batches_tracked", "stage4.0.branches.1.1.conv2.weight", "stage4.0.branches.1.1.bn2.weight", "stage4.0.branches.1.1.bn2.bias", "stage4.0.branches.1.1.bn2.running_mean", "stage4.0.branches.1.1.bn2.running_var", "stage4.0.branches.1.1.bn2.num_batches_tracked", "stage4.0.branches.1.2.conv1.weight", "stage4.0.branches.1.2.bn1.weight", "stage4.0.branches.1.2.bn1.bias", "stage4.0.branches.1.2.bn1.running_mean", "stage4.0.branches.1.2.bn1.running_var", "stage4.0.branches.1.2.bn1.num_batches_tracked", "stage4.0.branches.1.2.conv2.weight", "stage4.0.branches.1.2.bn2.weight", "stage4.0.branches.1.2.bn2.bias", "stage4.0.branches.1.2.bn2.running_mean", "stage4.0.branches.1.2.bn2.running_var", "stage4.0.branches.1.2.bn2.num_batches_tracked", "stage4.0.branches.1.3.conv1.weight", "stage4.0.branches.1.3.bn1.weight", "stage4.0.branches.1.3.bn1.bias", "stage4.0.branches.1.3.bn1.running_mean", "stage4.0.branches.1.3.bn1.running_var", "stage4.0.branches.1.3.bn1.num_batches_tracked", "stage4.0.branches.1.3.conv2.weight", "stage4.0.branches.1.3.bn2.weight", "stage4.0.branches.1.3.bn2.bias", "stage4.0.branches.1.3.bn2.running_mean", "stage4.0.branches.1.3.bn2.running_var", "stage4.0.branches.1.3.bn2.num_batches_tracked", "stage4.0.branches.2.0.conv1.weight", "stage4.0.branches.2.0.bn1.weight", "stage4.0.branches.2.0.bn1.bias", "stage4.0.branches.2.0.bn1.running_mean", "stage4.0.branches.2.0.bn1.running_var", "stage4.0.branches.2.0.bn1.num_batches_tracked", "stage4.0.branches.2.0.conv2.weight", "stage4.0.branches.2.0.bn2.weight", "stage4.0.branches.2.0.bn2.bias", "stage4.0.branches.2.0.bn2.running_mean", "stage4.0.branches.2.0.bn2.running_var", "stage4.0.branches.2.0.bn2.num_batches_tracked", "stage4.0.branches.2.1.conv1.weight", "stage4.0.branches.2.1.bn1.weight", "stage4.0.branches.2.1.bn1.bias", "stage4.0.branches.2.1.bn1.running_mean", "stage4.0.branches.2.1.bn1.running_var", "stage4.0.branches.2.1.bn1.num_batches_tracked", "stage4.0.branches.2.1.conv2.weight", "stage4.0.branches.2.1.bn2.weight", "stage4.0.branches.2.1.bn2.bias", "stage4.0.branches.2.1.bn2.running_mean", "stage4.0.branches.2.1.bn2.running_var", "stage4.0.branches.2.1.bn2.num_batches_tracked", "stage4.0.branches.2.2.conv1.weight", "stage4.0.branches.2.2.bn1.weight", "stage4.0.branches.2.2.bn1.bias", "stage4.0.branches.2.2.bn1.running_mean", "stage4.0.branches.2.2.bn1.running_var", "stage4.0.branches.2.2.bn1.num_batches_tracked", "stage4.0.branches.2.2.conv2.weight", "stage4.0.branches.2.2.bn2.weight", "stage4.0.branches.2.2.bn2.bias", "stage4.0.branches.2.2.bn2.running_mean", "stage4.0.branches.2.2.bn2.running_var", "stage4.0.branches.2.2.bn2.num_batches_tracked", "stage4.0.branches.2.3.conv1.weight", "stage4.0.branches.2.3.bn1.weight", "stage4.0.branches.2.3.bn1.bias", "stage4.0.branches.2.3.bn1.running_mean", "stage4.0.branches.2.3.bn1.running_var", "stage4.0.branches.2.3.bn1.num_batches_tracked", "stage4.0.branches.2.3.conv2.weight", "stage4.0.branches.2.3.bn2.weight", "stage4.0.branches.2.3.bn2.bias", "stage4.0.branches.2.3.bn2.running_mean", "stage4.0.branches.2.3.bn2.running_var", "stage4.0.branches.2.3.bn2.num_batches_tracked", "stage4.0.branches.3.0.conv1.weight", "stage4.0.branches.3.0.bn1.weight", "stage4.0.branches.3.0.bn1.bias", "stage4.0.branches.3.0.bn1.running_mean", "stage4.0.branches.3.0.bn1.running_var", "stage4.0.branches.3.0.bn1.num_batches_tracked", "stage4.0.branches.3.0.conv2.weight", "stage4.0.branches.3.0.bn2.weight", "stage4.0.branches.3.0.bn2.bias", "stage4.0.branches.3.0.bn2.running_mean", "stage4.0.branches.3.0.bn2.running_var", "stage4.0.branches.3.0.bn2.num_batches_tracked", "stage4.0.branches.3.1.conv1.weight", "stage4.0.branches.3.1.bn1.weight", "stage4.0.branches.3.1.bn1.bias", "stage4.0.branches.3.1.bn1.running_mean", "stage4.0.branches.3.1.bn1.running_var", "stage4.0.branches.3.1.bn1.num_batches_tracked", "stage4.0.branches.3.1.conv2.weight", "stage4.0.branches.3.1.bn2.weight", "stage4.0.branches.3.1.bn2.bias", "stage4.0.branches.3.1.bn2.running_mean", "stage4.0.branches.3.1.bn2.running_var", "stage4.0.branches.3.1.bn2.num_batches_tracked", "stage4.0.branches.3.2.conv1.weight", "stage4.0.branches.3.2.bn1.weight", "stage4.0.branches.3.2.bn1.bias", "stage4.0.branches.3.2.bn1.running_mean", "stage4.0.branches.3.2.bn1.running_var", "stage4.0.branches.3.2.bn1.num_batches_tracked", "stage4.0.branches.3.2.conv2.weight", "stage4.0.branches.3.2.bn2.weight", "stage4.0.branches.3.2.bn2.bias", "stage4.0.branches.3.2.bn2.running_mean", "stage4.0.branches.3.2.bn2.running_var", "stage4.0.branches.3.2.bn2.num_batches_tracked", "stage4.0.branches.3.3.conv1.weight", "stage4.0.branches.3.3.bn1.weight", "stage4.0.branches.3.3.bn1.bias", "stage4.0.branches.3.3.bn1.running_mean", "stage4.0.branches.3.3.bn1.running_var", "stage4.0.branches.3.3.bn1.num_batches_tracked", "stage4.0.branches.3.3.conv2.weight", "stage4.0.branches.3.3.bn2.weight", "stage4.0.branches.3.3.bn2.bias", "stage4.0.branches.3.3.bn2.running_mean", "stage4.0.branches.3.3.bn2.running_var", "stage4.0.branches.3.3.bn2.num_batches_tracked", "stage4.0.fuse_layers.0.1.0.weight", "stage4.0.fuse_layers.0.1.1.weight", "stage4.0.fuse_layers.0.1.1.bias", "stage4.0.fuse_layers.0.1.1.running_mean", "stage4.0.fuse_layers.0.1.1.running_var", "stage4.0.fuse_layers.0.1.1.num_batches_tracked", "stage4.0.fuse_layers.0.2.0.weight", "stage4.0.fuse_layers.0.2.1.weight", "stage4.0.fuse_layers.0.2.1.bias", "stage4.0.fuse_layers.0.2.1.running_mean", "stage4.0.fuse_layers.0.2.1.running_var", "stage4.0.fuse_layers.0.2.1.num_batches_tracked", "stage4.0.fuse_layers.0.3.0.weight", "stage4.0.fuse_layers.0.3.1.weight", "stage4.0.fuse_layers.0.3.1.bias", "stage4.0.fuse_layers.0.3.1.running_mean", "stage4.0.fuse_layers.0.3.1.running_var", "stage4.0.fuse_layers.0.3.1.num_batches_tracked", "stage4.0.fuse_layers.1.0.0.0.weight", "stage4.0.fuse_layers.1.0.0.1.weight", "stage4.0.fuse_layers.1.0.0.1.bias", "stage4.0.fuse_layers.1.0.0.1.running_mean", "stage4.0.fuse_layers.1.0.0.1.running_var", "stage4.0.fuse_layers.1.0.0.1.num_batches_tracked", "stage4.0.fuse_layers.1.2.0.weight", "stage4.0.fuse_layers.1.2.1.weight", "stage4.0.fuse_layers.1.2.1.bias", "stage4.0.fuse_layers.1.2.1.running_mean", "stage4.0.fuse_layers.1.2.1.running_var", "stage4.0.fuse_layers.1.2.1.num_batches_tracked", "stage4.0.fuse_layers.1.3.0.weight", "stage4.0.fuse_layers.1.3.1.weight", "stage4.0.fuse_layers.1.3.1.bias", "stage4.0.fuse_layers.1.3.1.running_mean", "stage4.0.fuse_layers.1.3.1.running_var", "stage4.0.fuse_layers.1.3.1.num_batches_tracked", "stage4.0.fuse_layers.2.0.0.0.weight", "stage4.0.fuse_layers.2.0.0.1.weight", "stage4.0.fuse_layers.2.0.0.1.bias", "stage4.0.fuse_layers.2.0.0.1.running_mean", "stage4.0.fuse_layers.2.0.0.1.running_var", "stage4.0.fuse_layers.2.0.0.1.num_batches_tracked", "stage4.0.fuse_layers.2.0.1.0.weight", "stage4.0.fuse_layers.2.0.1.1.weight", "stage4.0.fuse_layers.2.0.1.1.bias", "stage4.0.fuse_layers.2.0.1.1.running_mean", "stage4.0.fuse_layers.2.0.1.1.running_var", "stage4.0.fuse_layers.2.0.1.1.num_batches_tracked", "stage4.0.fuse_layers.2.1.0.0.weight", "stage4.0.fuse_layers.2.1.0.1.weight", "stage4.0.fuse_layers.2.1.0.1.bias", "stage4.0.fuse_layers.2.1.0.1.running_mean", "stage4.0.fuse_layers.2.1.0.1.running_var", "stage4.0.fuse_layers.2.1.0.1.num_batches_tracked", "stage4.0.fuse_layers.2.3.0.weight", "stage4.0.fuse_layers.2.3.1.weight", "stage4.0.fuse_layers.2.3.1.bias", "stage4.0.fuse_layers.2.3.1.running_mean", "stage4.0.fuse_layers.2.3.1.running_var", "stage4.0.fuse_layers.2.3.1.num_batches_tracked", "stage4.0.fuse_layers.3.0.0.0.weight", "stage4.0.fuse_layers.3.0.0.1.weight", "stage4.0.fuse_layers.3.0.0.1.bias", "stage4.0.fuse_layers.3.0.0.1.running_mean", "stage4.0.fuse_layers.3.0.0.1.running_var", "stage4.0.fuse_layers.3.0.0.1.num_batches_tracked", "stage4.0.fuse_layers.3.0.1.0.weight", "stage4.0.fuse_layers.3.0.1.1.weight", "stage4.0.fuse_layers.3.0.1.1.bias", "stage4.0.fuse_layers.3.0.1.1.running_mean", "stage4.0.fuse_layers.3.0.1.1.running_var", "stage4.0.fuse_layers.3.0.1.1.num_batches_tracked", "stage4.0.fuse_layers.3.0.2.0.weight", "stage4.0.fuse_layers.3.0.2.1.weight", "stage4.0.fuse_layers.3.0.2.1.bias", "stage4.0.fuse_layers.3.0.2.1.running_mean", "stage4.0.fuse_layers.3.0.2.1.running_var", "stage4.0.fuse_layers.3.0.2.1.num_batches_tracked", "stage4.0.fuse_layers.3.1.0.0.weight", "stage4.0.fuse_layers.3.1.0.1.weight", "stage4.0.fuse_layers.3.1.0.1.bias", "stage4.0.fuse_layers.3.1.0.1.running_mean", "stage4.0.fuse_layers.3.1.0.1.running_var", "stage4.0.fuse_layers.3.1.0.1.num_batches_tracked", "stage4.0.fuse_layers.3.1.1.0.weight", "stage4.0.fuse_layers.3.1.1.1.weight", "stage4.0.fuse_layers.3.1.1.1.bias", "stage4.0.fuse_layers.3.1.1.1.running_mean", "stage4.0.fuse_layers.3.1.1.1.running_var", "stage4.0.fuse_layers.3.1.1.1.num_batches_tracked", "stage4.0.fuse_layers.3.2.0.0.weight", "stage4.0.fuse_layers.3.2.0.1.weight", "stage4.0.fuse_layers.3.2.0.1.bias", "stage4.0.fuse_layers.3.2.0.1.running_mean", "stage4.0.fuse_layers.3.2.0.1.running_var", "stage4.0.fuse_layers.3.2.0.1.num_batches_tracked", "stage4.1.branches.0.0.conv1.weight", "stage4.1.branches.0.0.bn1.weight", "stage4.1.branches.0.0.bn1.bias", "stage4.1.branches.0.0.bn1.running_mean", "stage4.1.branches.0.0.bn1.running_var", "stage4.1.branches.0.0.bn1.num_batches_tracked", "stage4.1.branches.0.0.conv2.weight", "stage4.1.branches.0.0.bn2.weight", "stage4.1.branches.0.0.bn2.bias", "stage4.1.branches.0.0.bn2.running_mean", "stage4.1.branches.0.0.bn2.running_var", "stage4.1.branches.0.0.bn2.num_batches_tracked", "stage4.1.branches.0.1.conv1.weight", "stage4.1.branches.0.1.bn1.weight", "stage4.1.branches.0.1.bn1.bias", "stage4.1.branches.0.1.bn1.running_mean", "stage4.1.branches.0.1.bn1.running_var", "stage4.1.branches.0.1.bn1.num_batches_tracked", "stage4.1.branches.0.1.conv2.weight", "stage4.1.branches.0.1.bn2.weight", "stage4.1.branches.0.1.bn2.bias", "stage4.1.branches.0.1.bn2.running_mean", "stage4.1.branches.0.1.bn2.running_var", "stage4.1.branches.0.1.bn2.num_batches_tracked", "stage4.1.branches.0.2.conv1.weight", "stage4.1.branches.0.2.bn1.weight", "stage4.1.branches.0.2.bn1.bias", "stage4.1.branches.0.2.bn1.running_mean", "stage4.1.branches.0.2.bn1.running_var", "stage4.1.branches.0.2.bn1.num_batches_tracked", "stage4.1.branches.0.2.conv2.weight", "stage4.1.branches.0.2.bn2.weight", "stage4.1.branches.0.2.bn2.bias", "stage4.1.branches.0.2.bn2.running_mean", "stage4.1.branches.0.2.bn2.running_var", "stage4.1.branches.0.2.bn2.num_batches_tracked", "stage4.1.branches.0.3.conv1.weight", "stage4.1.branches.0.3.bn1.weight", "stage4.1.branches.0.3.bn1.bias", "stage4.1.branches.0.3.bn1.running_mean", "stage4.1.branches.0.3.bn1.running_var", "stage4.1.branches.0.3.bn1.num_batches_tracked", "stage4.1.branches.0.3.conv2.weight", "stage4.1.branches.0.3.bn2.weight", "stage4.1.branches.0.3.bn2.bias", "stage4.1.branches.0.3.bn2.running_mean", "stage4.1.branches.0.3.bn2.running_var", "stage4.1.branches.0.3.bn2.num_batches_tracked", "stage4.1.branches.1.0.conv1.weight", "stage4.1.branches.1.0.bn1.weight", "stage4.1.branches.1.0.bn1.bias", "stage4.1.branches.1.0.bn1.running_mean", "stage4.1.branches.1.0.bn1.running_var", "stage4.1.branches.1.0.bn1.num_batches_tracked", "stage4.1.branches.1.0.conv2.weight", "stage4.1.branches.1.0.bn2.weight", "stage4.1.branches.1.0.bn2.bias", "stage4.1.branches.1.0.bn2.running_mean", "stage4.1.branches.1.0.bn2.running_var", "stage4.1.branches.1.0.bn2.num_batches_tracked", "stage4.1.branches.1.1.conv1.weight", "stage4.1.branches.1.1.bn1.weight", "stage4.1.branches.1.1.bn1.bias", "stage4.1.branches.1.1.bn1.running_mean", "stage4.1.branches.1.1.bn1.running_var", "stage4.1.branches.1.1.bn1.num_batches_tracked", "stage4.1.branches.1.1.conv2.weight", "stage4.1.branches.1.1.bn2.weight", "stage4.1.branches.1.1.bn2.bias", "stage4.1.branches.1.1.bn2.running_mean", "stage4.1.branches.1.1.bn2.running_var", "stage4.1.branches.1.1.bn2.num_batches_tracked", "stage4.1.branches.1.2.conv1.weight", "stage4.1.branches.1.2.bn1.weight", "stage4.1.branches.1.2.bn1.bias", "stage4.1.branches.1.2.bn1.running_mean", "stage4.1.branches.1.2.bn1.running_var", "stage4.1.branches.1.2.bn1.num_batches_tracked", "stage4.1.branches.1.2.conv2.weight", "stage4.1.branches.1.2.bn2.weight", "stage4.1.branches.1.2.bn2.bias", "stage4.1.branches.1.2.bn2.running_mean", "stage4.1.branches.1.2.bn2.running_var", "stage4.1.branches.1.2.bn2.num_batches_tracked", "stage4.1.branches.1.3.conv1.weight", "stage4.1.branches.1.3.bn1.weight", "stage4.1.branches.1.3.bn1.bias", "stage4.1.branches.1.3.bn1.running_mean", "stage4.1.branches.1.3.bn1.running_var", "stage4.1.branches.1.3.bn1.num_batches_tracked", "stage4.1.branches.1.3.conv2.weight", "stage4.1.branches.1.3.bn2.weight", "stage4.1.branches.1.3.bn2.bias", "stage4.1.branches.1.3.bn2.running_mean", "stage4.1.branches.1.3.bn2.running_var", "stage4.1.branches.1.3.bn2.num_batches_tracked", "stage4.1.branches.2.0.conv1.weight", "stage4.1.branches.2.0.bn1.weight", "stage4.1.branches.2.0.bn1.bias", "stage4.1.branches.2.0.bn1.running_mean", "stage4.1.branches.2.0.bn1.running_var", "stage4.1.branches.2.0.bn1.num_batches_tracked", "stage4.1.branches.2.0.conv2.weight", "stage4.1.branches.2.0.bn2.weight", "stage4.1.branches.2.0.bn2.bias", "stage4.1.branches.2.0.bn2.running_mean", "stage4.1.branches.2.0.bn2.running_var", "stage4.1.branches.2.0.bn2.num_batches_tracked", "stage4.1.branches.2.1.conv1.weight", "stage4.1.branches.2.1.bn1.weight", "stage4.1.branches.2.1.bn1.bias", "stage4.1.branches.2.1.bn1.running_mean", "stage4.1.branches.2.1.bn1.running_var", "stage4.1.branches.2.1.bn1.num_batches_tracked", "stage4.1.branches.2.1.conv2.weight", "stage4.1.branches.2.1.bn2.weight", "stage4.1.branches.2.1.bn2.bias", "stage4.1.branches.2.1.bn2.running_mean", "stage4.1.branches.2.1.bn2.running_var", "stage4.1.branches.2.1.bn2.num_batches_tracked", "stage4.1.branches.2.2.conv1.weight", "stage4.1.branches.2.2.bn1.weight", "stage4.1.branches.2.2.bn1.bias", "stage4.1.branches.2.2.bn1.running_mean", "stage4.1.branches.2.2.bn1.running_var", "stage4.1.branches.2.2.bn1.num_batches_tracked", "stage4.1.branches.2.2.conv2.weight", "stage4.1.branches.2.2.bn2.weight", "stage4.1.branches.2.2.bn2.bias", "stage4.1.branches.2.2.bn2.running_mean", "stage4.1.branches.2.2.bn2.running_var", "stage4.1.branches.2.2.bn2.num_batches_tracked", "stage4.1.branches.2.3.conv1.weight", "stage4.1.branches.2.3.bn1.weight", "stage4.1.branches.2.3.bn1.bias", "stage4.1.branches.2.3.bn1.running_mean", "stage4.1.branches.2.3.bn1.running_var", "stage4.1.branches.2.3.bn1.num_batches_tracked", "stage4.1.branches.2.3.conv2.weight", "stage4.1.branches.2.3.bn2.weight", "stage4.1.branches.2.3.bn2.bias", "stage4.1.branches.2.3.bn2.running_mean", "stage4.1.branches.2.3.bn2.running_var", "stage4.1.branches.2.3.bn2.num_batches_tracked", "stage4.1.branches.3.0.conv1.weight", "stage4.1.branches.3.0.bn1.weight", "stage4.1.branches.3.0.bn1.bias", "stage4.1.branches.3.0.bn1.running_mean", "stage4.1.branches.3.0.bn1.running_var", "stage4.1.branches.3.0.bn1.num_batches_tracked", "stage4.1.branches.3.0.conv2.weight", "stage4.1.branches.3.0.bn2.weight", "stage4.1.branches.3.0.bn2.bias", "stage4.1.branches.3.0.bn2.running_mean", "stage4.1.branches.3.0.bn2.running_var", "stage4.1.branches.3.0.bn2.num_batches_tracked", "stage4.1.branches.3.1.conv1.weight", "stage4.1.branches.3.1.bn1.weight", "stage4.1.branches.3.1.bn1.bias", "stage4.1.branches.3.1.bn1.running_mean", "stage4.1.branches.3.1.bn1.running_var", "stage4.1.branches.3.1.bn1.num_batches_tracked", "stage4.1.branches.3.1.conv2.weight", "stage4.1.branches.3.1.bn2.weight", "stage4.1.branches.3.1.bn2.bias", "stage4.1.branches.3.1.bn2.running_mean", "stage4.1.branches.3.1.bn2.running_var", "stage4.1.branches.3.1.bn2.num_batches_tracked", "stage4.1.branches.3.2.conv1.weight", "stage4.1.branches.3.2.bn1.weight", "stage4.1.branches.3.2.bn1.bias", "stage4.1.branches.3.2.bn1.running_mean", "stage4.1.branches.3.2.bn1.running_var", "stage4.1.branches.3.2.bn1.num_batches_tracked", "stage4.1.branches.3.2.conv2.weight", "stage4.1.branches.3.2.bn2.weight", "stage4.1.branches.3.2.bn2.bias", "stage4.1.branches.3.2.bn2.running_mean", "stage4.1.branches.3.2.bn2.running_var", "stage4.1.branches.3.2.bn2.num_batches_tracked", "stage4.1.branches.3.3.conv1.weight", "stage4.1.branches.3.3.bn1.weight", "stage4.1.branches.3.3.bn1.bias", "stage4.1.branches.3.3.bn1.running_mean", "stage4.1.branches.3.3.bn1.running_var", "stage4.1.branches.3.3.bn1.num_batches_tracked", "stage4.1.branches.3.3.conv2.weight", "stage4.1.branches.3.3.bn2.weight", "stage4.1.branches.3.3.bn2.bias", "stage4.1.branches.3.3.bn2.running_mean", "stage4.1.branches.3.3.bn2.running_var", "stage4.1.branches.3.3.bn2.num_batches_tracked", "stage4.1.fuse_layers.0.1.0.weight", "stage4.1.fuse_layers.0.1.1.weight", "stage4.1.fuse_layers.0.1.1.bias", "stage4.1.fuse_layers.0.1.1.running_mean", "stage4.1.fuse_layers.0.1.1.running_var", "stage4.1.fuse_layers.0.1.1.num_batches_tracked", "stage4.1.fuse_layers.0.2.0.weight", "stage4.1.fuse_layers.0.2.1.weight", "stage4.1.fuse_layers.0.2.1.bias", "stage4.1.fuse_layers.0.2.1.running_mean", "stage4.1.fuse_layers.0.2.1.running_var", "stage4.1.fuse_layers.0.2.1.num_batches_tracked", "stage4.1.fuse_layers.0.3.0.weight", "stage4.1.fuse_layers.0.3.1.weight", "stage4.1.fuse_layers.0.3.1.bias", "stage4.1.fuse_layers.0.3.1.running_mean", "stage4.1.fuse_layers.0.3.1.running_var", "stage4.1.fuse_layers.0.3.1.num_batches_tracked", "stage4.1.fuse_layers.1.0.0.0.weight", "stage4.1.fuse_layers.1.0.0.1.weight", "stage4.1.fuse_layers.1.0.0.1.bias", "stage4.1.fuse_layers.1.0.0.1.running_mean", "stage4.1.fuse_layers.1.0.0.1.running_var", "stage4.1.fuse_layers.1.0.0.1.num_batches_tracked", "stage4.1.fuse_layers.1.2.0.weight", "stage4.1.fuse_layers.1.2.1.weight", "stage4.1.fuse_layers.1.2.1.bias", "stage4.1.fuse_layers.1.2.1.running_mean", "stage4.1.fuse_layers.1.2.1.running_var", "stage4.1.fuse_layers.1.2.1.num_batches_tracked", "stage4.1.fuse_layers.1.3.0.weight", "stage4.1.fuse_layers.1.3.1.weight", "stage4.1.fuse_layers.1.3.1.bias", "stage4.1.fuse_layers.1.3.1.running_mean", "stage4.1.fuse_layers.1.3.1.running_var", "stage4.1.fuse_layers.1.3.1.num_batches_tracked", "stage4.1.fuse_layers.2.0.0.0.weight", "stage4.1.fuse_layers.2.0.0.1.weight", "stage4.1.fuse_layers.2.0.0.1.bias", "stage4.1.fuse_layers.2.0.0.1.running_mean", "stage4.1.fuse_layers.2.0.0.1.running_var", "stage4.1.fuse_layers.2.0.0.1.num_batches_tracked", "stage4.1.fuse_layers.2.0.1.0.weight", "stage4.1.fuse_layers.2.0.1.1.weight", "stage4.1.fuse_layers.2.0.1.1.bias", "stage4.1.fuse_layers.2.0.1.1.running_mean", "stage4.1.fuse_layers.2.0.1.1.running_var", "stage4.1.fuse_layers.2.0.1.1.num_batches_tracked", "stage4.1.fuse_layers.2.1.0.0.weight", "stage4.1.fuse_layers.2.1.0.1.weight", "stage4.1.fuse_layers.2.1.0.1.bias", "stage4.1.fuse_layers.2.1.0.1.running_mean", "stage4.1.fuse_layers.2.1.0.1.running_var", "stage4.1.fuse_layers.2.1.0.1.num_batches_tracked", "stage4.1.fuse_layers.2.3.0.weight", "stage4.1.fuse_layers.2.3.1.weight", "stage4.1.fuse_layers.2.3.1.bias", "stage4.1.fuse_layers.2.3.1.running_mean", "stage4.1.fuse_layers.2.3.1.running_var", "stage4.1.fuse_layers.2.3.1.num_batches_tracked", "stage4.1.fuse_layers.3.0.0.0.weight", "stage4.1.fuse_layers.3.0.0.1.weight", "stage4.1.fuse_layers.3.0.0.1.bias", "stage4.1.fuse_layers.3.0.0.1.running_mean", "stage4.1.fuse_layers.3.0.0.1.running_var", "stage4.1.fuse_layers.3.0.0.1.num_batches_tracked", "stage4.1.fuse_layers.3.0.1.0.weight", "stage4.1.fuse_layers.3.0.1.1.weight", "stage4.1.fuse_layers.3.0.1.1.bias", "stage4.1.fuse_layers.3.0.1.1.running_mean", "stage4.1.fuse_layers.3.0.1.1.running_var", "stage4.1.fuse_layers.3.0.1.1.num_batches_tracked", "stage4.1.fuse_layers.3.0.2.0.weight", "stage4.1.fuse_layers.3.0.2.1.weight", "stage4.1.fuse_layers.3.0.2.1.bias", "stage4.1.fuse_layers.3.0.2.1.running_mean", "stage4.1.fuse_layers.3.0.2.1.running_var", "stage4.1.fuse_layers.3.0.2.1.num_batches_tracked", "stage4.1.fuse_layers.3.1.0.0.weight", "stage4.1.fuse_layers.3.1.0.1.weight", "stage4.1.fuse_layers.3.1.0.1.bias", "stage4.1.fuse_layers.3.1.0.1.running_mean", "stage4.1.fuse_layers.3.1.0.1.running_var", "stage4.1.fuse_layers.3.1.0.1.num_batches_tracked", "stage4.1.fuse_layers.3.1.1.0.weight", "stage4.1.fuse_layers.3.1.1.1.weight", "stage4.1.fuse_layers.3.1.1.1.bias", "stage4.1.fuse_layers.3.1.1.1.running_mean", "stage4.1.fuse_layers.3.1.1.1.running_var", "stage4.1.fuse_layers.3.1.1.1.num_batches_tracked", "stage4.1.fuse_layers.3.2.0.0.weight", "stage4.1.fuse_layers.3.2.0.1.weight", "stage4.1.fuse_layers.3.2.0.1.bias", "stage4.1.fuse_layers.3.2.0.1.running_mean", "stage4.1.fuse_layers.3.2.0.1.running_var", "stage4.1.fuse_layers.3.2.0.1.num_batches_tracked", "stage4.2.branches.0.0.conv1.weight", "stage4.2.branches.0.0.bn1.weight", "stage4.2.branches.0.0.bn1.bias", "stage4.2.branches.0.0.bn1.running_mean", "stage4.2.branches.0.0.bn1.running_var", "stage4.2.branches.0.0.bn1.num_batches_tracked", "stage4.2.branches.0.0.conv2.weight", "stage4.2.branches.0.0.bn2.weight", "stage4.2.branches.0.0.bn2.bias", "stage4.2.branches.0.0.bn2.running_mean", "stage4.2.branches.0.0.bn2.running_var", "stage4.2.branches.0.0.bn2.num_batches_tracked", "stage4.2.branches.0.1.conv1.weight", "stage4.2.branches.0.1.bn1.weight", "stage4.2.branches.0.1.bn1.bias", "stage4.2.branches.0.1.bn1.running_mean", "stage4.2.branches.0.1.bn1.running_var", "stage4.2.branches.0.1.bn1.num_batches_tracked", "stage4.2.branches.0.1.conv2.weight", "stage4.2.branches.0.1.bn2.weight", "stage4.2.branches.0.1.bn2.bias", "stage4.2.branches.0.1.bn2.running_mean", "stage4.2.branches.0.1.bn2.running_var", "stage4.2.branches.0.1.bn2.num_batches_tracked", "stage4.2.branches.0.2.conv1.weight", "stage4.2.branches.0.2.bn1.weight", "stage4.2.branches.0.2.bn1.bias", "stage4.2.branches.0.2.bn1.running_mean", "stage4.2.branches.0.2.bn1.running_var", "stage4.2.branches.0.2.bn1.num_batches_tracked", "stage4.2.branches.0.2.conv2.weight", "stage4.2.branches.0.2.bn2.weight", "stage4.2.branches.0.2.bn2.bias", "stage4.2.branches.0.2.bn2.running_mean", "stage4.2.branches.0.2.bn2.running_var", "stage4.2.branches.0.2.bn2.num_batches_tracked", "stage4.2.branches.0.3.conv1.weight", "stage4.2.branches.0.3.bn1.weight", "stage4.2.branches.0.3.bn1.bias", "stage4.2.branches.0.3.bn1.running_mean", "stage4.2.branches.0.3.bn1.running_var", "stage4.2.branches.0.3.bn1.num_batches_tracked", "stage4.2.branches.0.3.conv2.weight", "stage4.2.branches.0.3.bn2.weight", "stage4.2.branches.0.3.bn2.bias", "stage4.2.branches.0.3.bn2.running_mean", "stage4.2.branches.0.3.bn2.running_var", "stage4.2.branches.0.3.bn2.num_batches_tracked", "stage4.2.branches.1.0.conv1.weight", "stage4.2.branches.1.0.bn1.weight", "stage4.2.branches.1.0.bn1.bias", "stage4.2.branches.1.0.bn1.running_mean", "stage4.2.branches.1.0.bn1.running_var", "stage4.2.branches.1.0.bn1.num_batches_tracked", "stage4.2.branches.1.0.conv2.weight", "stage4.2.branches.1.0.bn2.weight", "stage4.2.branches.1.0.bn2.bias", "stage4.2.branches.1.0.bn2.running_mean", "stage4.2.branches.1.0.bn2.running_var", "stage4.2.branches.1.0.bn2.num_batches_tracked", "stage4.2.branches.1.1.conv1.weight", "stage4.2.branches.1.1.bn1.weight", "stage4.2.branches.1.1.bn1.bias", "stage4.2.branches.1.1.bn1.running_mean", "stage4.2.branches.1.1.bn1.running_var", "stage4.2.branches.1.1.bn1.num_batches_tracked", "stage4.2.branches.1.1.conv2.weight", "stage4.2.branches.1.1.bn2.weight", "stage4.2.branches.1.1.bn2.bias", "stage4.2.branches.1.1.bn2.running_mean", "stage4.2.branches.1.1.bn2.running_var", "stage4.2.branches.1.1.bn2.num_batches_tracked", "stage4.2.branches.1.2.conv1.weight", "stage4.2.branches.1.2.bn1.weight", "stage4.2.branches.1.2.bn1.bias", "stage4.2.branches.1.2.bn1.running_mean", "stage4.2.branches.1.2.bn1.running_var", "stage4.2.branches.1.2.bn1.num_batches_tracked", "stage4.2.branches.1.2.conv2.weight", "stage4.2.branches.1.2.bn2.weight", "stage4.2.branches.1.2.bn2.bias", "stage4.2.branches.1.2.bn2.running_mean", "stage4.2.branches.1.2.bn2.running_var", "stage4.2.branches.1.2.bn2.num_batches_tracked", "stage4.2.branches.1.3.conv1.weight", "stage4.2.branches.1.3.bn1.weight", "stage4.2.branches.1.3.bn1.bias", "stage4.2.branches.1.3.bn1.running_mean", "stage4.2.branches.1.3.bn1.running_var", "stage4.2.branches.1.3.bn1.num_batches_tracked", "stage4.2.branches.1.3.conv2.weight", "stage4.2.branches.1.3.bn2.weight", "stage4.2.branches.1.3.bn2.bias", "stage4.2.branches.1.3.bn2.running_mean", "stage4.2.branches.1.3.bn2.running_var", "stage4.2.branches.1.3.bn2.num_batches_tracked", "stage4.2.branches.2.0.conv1.weight", "stage4.2.branches.2.0.bn1.weight", "stage4.2.branches.2.0.bn1.bias", "stage4.2.branches.2.0.bn1.running_mean", "stage4.2.branches.2.0.bn1.running_var", "stage4.2.branches.2.0.bn1.num_batches_tracked", "stage4.2.branches.2.0.conv2.weight", "stage4.2.branches.2.0.bn2.weight", "stage4.2.branches.2.0.bn2.bias", "stage4.2.branches.2.0.bn2.running_mean", "stage4.2.branches.2.0.bn2.running_var", "stage4.2.branches.2.0.bn2.num_batches_tracked", "stage4.2.branches.2.1.conv1.weight", "stage4.2.branches.2.1.bn1.weight", "stage4.2.branches.2.1.bn1.bias", "stage4.2.branches.2.1.bn1.running_mean", "stage4.2.branches.2.1.bn1.running_var", "stage4.2.branches.2.1.bn1.num_batches_tracked", "stage4.2.branches.2.1.conv2.weight", "stage4.2.branches.2.1.bn2.weight", "stage4.2.branches.2.1.bn2.bias", "stage4.2.branches.2.1.bn2.running_mean", "stage4.2.branches.2.1.bn2.running_var", "stage4.2.branches.2.1.bn2.num_batches_tracked", "stage4.2.branches.2.2.conv1.weight", "stage4.2.branches.2.2.bn1.weight", "stage4.2.branches.2.2.bn1.bias", "stage4.2.branches.2.2.bn1.running_mean", "stage4.2.branches.2.2.bn1.running_var", "stage4.2.branches.2.2.bn1.num_batches_tracked", "stage4.2.branches.2.2.conv2.weight", "stage4.2.branches.2.2.bn2.weight", "stage4.2.branches.2.2.bn2.bias", "stage4.2.branches.2.2.bn2.running_mean", "stage4.2.branches.2.2.bn2.running_var", "stage4.2.branches.2.2.bn2.num_batches_tracked", "stage4.2.branches.2.3.conv1.weight", "stage4.2.branches.2.3.bn1.weight", "stage4.2.branches.2.3.bn1.bias", "stage4.2.branches.2.3.bn1.running_mean", "stage4.2.branches.2.3.bn1.running_var", "stage4.2.branches.2.3.bn1.num_batches_tracked", "stage4.2.branches.2.3.conv2.weight", "stage4.2.branches.2.3.bn2.weight", "stage4.2.branches.2.3.bn2.bias", "stage4.2.branches.2.3.bn2.running_mean", "stage4.2.branches.2.3.bn2.running_var", "stage4.2.branches.2.3.bn2.num_batches_tracked", "stage4.2.branches.3.0.conv1.weight", "stage4.2.branches.3.0.bn1.weight", "stage4.2.branches.3.0.bn1.bias", "stage4.2.branches.3.0.bn1.running_mean", "stage4.2.branches.3.0.bn1.running_var", "stage4.2.branches.3.0.bn1.num_batches_tracked", "stage4.2.branches.3.0.conv2.weight", "stage4.2.branches.3.0.bn2.weight", "stage4.2.branches.3.0.bn2.bias", "stage4.2.branches.3.0.bn2.running_mean", "stage4.2.branches.3.0.bn2.running_var", "stage4.2.branches.3.0.bn2.num_batches_tracked", "stage4.2.branches.3.1.conv1.weight", "stage4.2.branches.3.1.bn1.weight", "stage4.2.branches.3.1.bn1.bias", "stage4.2.branches.3.1.bn1.running_mean", "stage4.2.branches.3.1.bn1.running_var", "stage4.2.branches.3.1.bn1.num_batches_tracked", "stage4.2.branches.3.1.conv2.weight", "stage4.2.branches.3.1.bn2.weight", "stage4.2.branches.3.1.bn2.bias", "stage4.2.branches.3.1.bn2.running_mean", "stage4.2.branches.3.1.bn2.running_var", "stage4.2.branches.3.1.bn2.num_batches_tracked", "stage4.2.branches.3.2.conv1.weight", "stage4.2.branches.3.2.bn1.weight", "stage4.2.branches.3.2.bn1.bias", "stage4.2.branches.3.2.bn1.running_mean", "stage4.2.branches.3.2.bn1.running_var", "stage4.2.branches.3.2.bn1.num_batches_tracked", "stage4.2.branches.3.2.conv2.weight", "stage4.2.branches.3.2.bn2.weight", "stage4.2.branches.3.2.bn2.bias", "stage4.2.branches.3.2.bn2.running_mean", "stage4.2.branches.3.2.bn2.running_var", "stage4.2.branches.3.2.bn2.num_batches_tracked", "stage4.2.branches.3.3.conv1.weight", "stage4.2.branches.3.3.bn1.weight", "stage4.2.branches.3.3.bn1.bias", "stage4.2.branches.3.3.bn1.running_mean", "stage4.2.branches.3.3.bn1.running_var", "stage4.2.branches.3.3.bn1.num_batches_tracked", "stage4.2.branches.3.3.conv2.weight", "stage4.2.branches.3.3.bn2.weight", "stage4.2.branches.3.3.bn2.bias", "stage4.2.branches.3.3.bn2.running_mean", "stage4.2.branches.3.3.bn2.running_var", "stage4.2.branches.3.3.bn2.num_batches_tracked", "stage4.2.fuse_layers.0.1.0.weight", "stage4.2.fuse_layers.0.1.1.weight", "stage4.2.fuse_layers.0.1.1.bias", "stage4.2.fuse_layers.0.1.1.running_mean", "stage4.2.fuse_layers.0.1.1.running_var", "stage4.2.fuse_layers.0.1.1.num_batches_tracked", "stage4.2.fuse_layers.0.2.0.weight", "stage4.2.fuse_layers.0.2.1.weight", "stage4.2.fuse_layers.0.2.1.bias", "stage4.2.fuse_layers.0.2.1.running_mean", "stage4.2.fuse_layers.0.2.1.running_var", "stage4.2.fuse_layers.0.2.1.num_batches_tracked", "stage4.2.fuse_layers.0.3.0.weight", "stage4.2.fuse_layers.0.3.1.weight", "stage4.2.fuse_layers.0.3.1.bias", "stage4.2.fuse_layers.0.3.1.running_mean", "stage4.2.fuse_layers.0.3.1.running_var", "stage4.2.fuse_layers.0.3.1.num_batches_tracked", "layer1.3.conv1.weight", "layer1.3.bn1.weight", "layer1.3.bn1.bias", "layer1.3.bn1.running_mean", "layer1.3.bn1.running_var", "layer1.3.bn1.num_batches_tracked", "layer1.3.conv2.weight", "layer1.3.bn2.weight", "layer1.3.bn2.bias", "layer1.3.bn2.running_mean", "layer1.3.bn2.running_var", "layer1.3.bn2.num_batches_tracked", "layer1.3.conv3.weight", "layer1.3.bn3.weight", "layer1.3.bn3.bias", "layer1.3.bn3.running_mean", "layer1.3.bn3.running_var", "layer1.3.bn3.num_batches_tracked". 
	size mismatch for conv1.weight: copying a param with shape torch.Size([64, 3, 3, 3]) from checkpoint, the shape in current model is torch.Size([64, 3, 7, 7]).
	size mismatch for final_layer.weight: copying a param with shape torch.Size([17, 48, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 256, 1, 1]).
	size mismatch for final_layer.bias: copying a param with shape torch.Size([17]) from checkpoint, the shape in current model is torch.Size([16]).

Is there a config.py file that can work for this repo? If so, could you please upload?

EDIT:
I used the following config and frozen model:

model_file = '/deep-high-resolution-net.pytorch/models/pytorch/pose_coco/pose_hrnet_w48_384x288.pth'
config_path = "/deep-high-resolution-net.pytorch/experiments/coco/hrnet/w48_384x288_adam_lr1e-3.yaml"

mpii submission

hi, i use test.json to creat a [7247. 16. 2] pred.mat, can i submitt pred.mat to mpii official website to have a test result?

core/loss.py confict modules/loss.py

class JointsMSELoss select "mean" as criterion,but in MSELoss just can select "none", "elementwise_mean","sum","error value"

How to find feature is not aligned.

@leoxiaobin Thanks for sharing your work.

Could you please explain how to find feature is not aligned.

SHIFT_HEATMAP operation in function.py

# feature is not aligned, shift flipped heatmap for higher accuracy
if config.TEST.SHIFT_HEATMAP:
    output_flipped[:, :, :, 2:] = \
        output_flipped.clone()[:, :, :, 0:-2]

Test coco valid dataset in w32_256x192 model (gt bbox) , and results as follow:

Arch	AP	Ap .5	AP .75	AP (M)	AP (L)	AR	AR .5	AR .75	AR (M)	AR (L)
NOT SHIFT_HEATMAP	0.720	0.932	0.802	0.692	0.771	0.774	0.946	0.850	0.736	0.832
SHIFT_HEATMAP 1 pixel	0.740	0.932	0.815	0.717	0.780	0.791	0.947	0.857	0.760	0.839
SHIFT_HEATMAP 2 pixel	0.674	0.928	0.779	0.661	0.703	0.738	0.942	0.830	0.713	0.775

Thank you so much!

how to make this model working in realtime(webcam)

How to get person detection?

First of all thank you for your excellent work. I have a question regarding person detection. In your paper it is mentioned that you use a person detector before feeding its output to the HRNet. Am I supposed download this separately and then feed its output to the HRNet? If so, what do the dataloaders in train and test.py do? Would it be possible for you to tell me which person detector has been used?

Getting pose estimation on batch of images

I am new to this but have used several pose estimators before, such as OpenPose and AlphaPose.

I was wondering whether there is a way to run a batch of images through the trained network, and receive the estimated pose from each image as an output.
In other pose estimators, this is done by saving as a .json with the keypoints given by coordinates.

Is this possible yet with this pose estimator?

Many thanks,
spoonerj

Hrnet slow compared to simple baseline

Hello,

Thank you very much for the implementation and the trained models.

I compared pose estimation run time of both pose_hrnet_w32_256x192.pth and pose_resnet_50_256x192.pth on the same large data-set I have (~5700 images with ~2 people in an image). In both runs I start measuring after 100 iterations to account for GPU warm-up. I measure run time without the detection part, yet including pre-proccessing of cropping and resizing the detection crops, which is done in the same way for both networks.

I get ~101 fps for pose_resnet_50_256x192, compared to ~82 fps for pose_hrnet_w32_256x192 on a single 2080ti gpu.

Could it be that hrnet is slower than simple baseline or am I missing something?

Question on nms

some question about person detection

Thanks for your wonderful work, but I have some question about person detection part.
As we know, for the top-down method, person detection result is a keypoint that determines the final keypoints performance.
In the paper, It just mentions using simple baseline's detection method, but in simple baseline's repo didn't contain anyting about this part.
For me, I have tried many detection method, but never achieved the AP of 60.9 in COCO test-dev2017.
May I ask which detection model you use and is there any plan to release detection model?

why set TEST.USE_GT_BBOX False

you provide coco test command

python tools/test.py \
    --cfg experiments/coco/hrnet/w32_256x192_adam_lr1e-3.yaml \
    TEST.MODEL_FILE models/pytorch/pose_coco/pose_hrnet_w32_256x192.pth \
    TEST.USE_GT_BBOX False

But I found I can get high accuracy only if I set it true. Otherwise, the accuracy is almost zero.

'center' and 'scale' in MPII annotation

Hi!
How can I get the 'center' and 'scale' in MPII annotation?
Are these values obtained by a detector?
Or they're just the average of the joints coordinates?

Train/Val split

Thanks for making the code available!
Could you explain how you split the data in MPII's "annot"? Is this data split universally agreed or just for this paper?

There are:
22246 individuals in "train.json",
29116 individuals in "trainval.json",
2958 individuals in "valid.json",
7247 individuals in "test.json".

What confuses me is that 22246 + 2958 != 29116. Also trainval.json is not used anywhere.

I'm wondering when we report results (like what you did for Table 3 in your paper), which split should we use for training and testing, respectively?

Where is the exchange unit across stages?

I can't find the exchange unit across stages in code, could you point that?
thank you.

Flip problem in training process

In training progress, flip augmentation need to shift 1 pixel after flipping the keypoints:
`def fliplr_joints(joints, joints_vis, width, matched_parts):
"""
flip coords
"""
# Flip horizontal
joints[:, 0] = width - joints[:, 0] - 1

# Change left-right parts
for pair in matched_parts:
    joints[pair[0], :], joints[pair[1], :] = \
        joints[pair[1], :], joints[pair[0], :].copy()
    joints_vis[pair[0], :], joints_vis[pair[1], :] = \
        joints_vis[pair[1], :], joints_vis[pair[0], :].copy()

return joints*joints_vis, joints_vis

`
I wonder why it need to shift 1 pixel? joints[:, 0] = width - joints[:, 0] - 1

opencv error

Thanks for sharing the code. When I tried to run the test commands both failed with
cv2.error: OpenCV(3.4.1) /io/opencv/modules/imgproc/src/color.cpp:11115: error: (-215) scn == 3 || scn == 4 in function cvtColor. Am I missing something? Thanks,

No pose_hrnet_w48_256x256.pth in GoogleDrive or OneDrive link?

heyyyyy, appreciate ur great work.

Here is my problem, I couldn't see your pre-trained model, models/pytorch/pose_mpii/pose_hrnet_w48_256x256.pth in neither Google Drive or One Drive link.

Will you upload it again? Or where can I download it? Or something else?

With this problem, it will come out below error message in the part Testing on COCO val2017 dataset using model zoo's models

FileNotFoundError: [Errno 2] No such file or directory: 'data/coco/annotations/person_keypoints_val2017.json'

Thanks!

person detection

How to generate the person detection result of COCO val2017 and test-dev2017 by myself?

What size is accepted by the pretrained COCO model?

I want to inference on a 1280x720 image. I got RuntimeError: The size of tensor a (180) must match the size of tensor b (184) at non-singleton dimension 3.

Why heatmap size is also downsample into 4 times smaller

Hi, may I ask why heatmap size is also 4x smaller than input image? I mean if you are keeping high resolution feature maps all the time, why do not just generate heatmaps with original input size for training and inference?
Is there something that I am missing?

How long is the training time for the network?

The performance is very impressive. But I wonder how long is the training time. The paper said that the training process is terminated within 210 epochs. I wonder how long is one epoch in 4 NVIDIA P100 GPU cards? Thanks.

Have you tested to use hrnet to object detect?

Thanks for your works! Have you tested to use hrnet to object detect? And what's the performance?

How do you deal with incomplete person appeared in image？network can not return right pose!

Sorry , something wrong,i can not process image.
For example,the whole image，only have one person upper body！So ，the detector will give a box [0,0,w,h].
in this project , the network use input image with " center of image == person center,in mpii"!
but in this kind of image,we only have upper body,we can not build this kind of input in inference time.

So, anyone can help me to deal with this kind of problem?^_^

Key

Would it be possible to run this in Windows 10?

HRNetV2-W18

Do you plan to release HRNetV2-w18 for this task?

Usage on CPU

Currently when trying to make libs on my pc(without gpu),I get the following error:

Could you give me some suggestions of it？Thank you！

A problem occur while running test.py

First I didn't have four GPUs, so I changed _C. WORKERS to 1 and changed model = torch. nn. DataParallel (model, device_ids = cfg. GPUS). cuda () to model = torch. nn. Parallel (model, device_ids= (0,). cuda (). When I run test.py, I found that my CPU was full (the CPU memory is 8G in total) and could not run. I got stuck inloading samples.Is there any way to solve this problem?

cant't download pretrained mode from the GoogleDrive and OneDrive

who can share the pretrained mode with the BAIDU Drive(百度网盘）， very thanks to you !

ImportError: gpu_nms.cpython-36m-x86_64-linux-gnu.so: undefined symbol: __cudaPopCallConfiguration

when i run "python tools/test.py --cfg experiments/coco/hrnet/w32_256x192_adam_lr1e-3.yaml TEST.MODEL_FILE models/pytorch/pose_coco/pose_hrnet_w32_256x192.pth TEST.USE_GT_BBOX False", i get such an error:

Traceback (most recent call last):
  File "tools/test.py", line 31, in <module>
    import dataset
  File "/data1/home/github/deep-high-resolution-net.pytorch/tools/../lib/dataset/__init__.py", line 12, in <module>
    from .coco import COCODataset as coco
  File "/data1/home/github/deep-high-resolution-net.pytorch/tools/../lib/dataset/coco.py", line 22, in <module>
    from nms.nms import oks_nms
  File "/data1/home/github/deep-high-resolution-net.pytorch/tools/../lib/nms/nms.py", line 14, in <module>
    from .gpu_nms import gpu_nms
ImportError: /data1/home/github/deep-high-resolution-net.pytorch/tools/../lib/nms/gpu_nms.cpython-36m-x86_64-linux-gnu.so: undefined symbol: __cudaPopCallConfiguration

This is my environment package configuration：

Package                    Version    
---------------         -----------
ca-certificates           2019.3.9
certifi                   2019.3.9 
cffi                      1.12.2
cloudpickle               0.8.1
cudatoolkit               9.0 
cycler                    0.10.0 
cython                    0.29.6 
dask                      1.1.4 
decorator                 4.4.0
easydict                  1.7
intel-openmp              2019.1
json-tricks               3.13.1
kiwisolver                1.0.1
libblas                   3.8.0
libcblas                  3.8.0
libffi                    3.2.1
libgcc-ng                 7.3.0
libgfortran-ng            7.2.0
liblapack                 3.8.0 
libstdcxx-ng              7.3.0
matplotlib                3.0.3 
mkl                       2019.1
ncurses                   6.1
networkx                  2.2 
ninja                     1.9.0
numpy                     1.16.2
openblas                  0.3.5
opencv-python             3.4.1.15
openssl                   1.1.1b
pandas                    0.24.2
pillow                    5.4.1
pip                       19.0.3
protobuf                  3.7.1
pycocotools               2.0 
pycparser                 2.19
pyparsing                 2.3.1
python                    3.6.7
python-dateutil           2.8.0
pytorch-nightly           1.0.0.dev20190326
pytz                      2018.9
pywavelets                1.0.2
pyyaml                    5.1
readline                  7.0
scikit-image              0.14.2pi
scipy                     1.2.1
setuptools                40.8.0
shapely                   1.6.4
six                       1.12.0 
sqlite                    3.26.0
tensorboardx              1.6
tk                        8.6.9
toolz                     0.9.0
torchvision               0.2.2.post3
wheel                     0.33.1
xz                        5.2.4
yacs                      0.1.6 
zlib                      1.2.11

Does anybody have the same problem?

Demo

First of all thank you for the great work. I'm currently trying to set up a demo of the estimator but run into some issues in the post-processing stage (the network output is a B x 17 x 128 x 128 for 512x512 images)
Are planning to release any helper functions for post-processing the output to a key-points ?

Many thanks

Does code include pose tracking as well?

I'm curious if the multi person tracking code is also in this repo?
Thanks

[BUILD ERROR] nms_kernel.cu: No such file or directory

(venv) juggernaut@xmen9:~/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/lib$ make
cd nms; python setup_linux.py build_ext --inplace; rm -rf build; cd ../../
running build_ext
cythoning cpu_nms.pyx to cpu_nms.c
/home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/venv/lib/python3.5/site-packages/Cython/Compiler/Main.py:367: FutureWarning: Cython directive 'language_level' not set, using 2 for now (Py2). This will change in a later release! File: /home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/lib/nms/cpu_nms.pyx
  tree = Parsing.p_module(s, pxd, full_module_name)
building 'cpu_nms' extension
creating build
creating build/temp.linux-x86_64-3.5
x86_64-linux-gnu-gcc -pthread -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -fPIC -I/home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/venv/lib/python3.5/site-packages/numpy/core/include -I/home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/venv/include -I/usr/include/python3.5m -c cpu_nms.c -o build/temp.linux-x86_64-3.5/cpu_nms.o -Wno-cpp -Wno-unused-function
x86_64-linux-gnu-gcc -pthread -shared -Wl,-O1 -Wl,-Bsymbolic-functions -Wl,-Bsymbolic-functions -Wl,-z,relro -Wl,-Bsymbolic-functions -Wl,-z,relro -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 build/temp.linux-x86_64-3.5/cpu_nms.o -o /home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/lib/nms/cpu_nms.cpython-35m-x86_64-linux-gnu.so
cythoning gpu_nms.pyx to gpu_nms.cpp
/home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/venv/lib/python3.5/site-packages/Cython/Compiler/Main.py:367: FutureWarning: Cython directive 'language_level' not set, using 2 for now (Py2). This will change in a later release! File: /home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/lib/nms/gpu_nms.pyx
  tree = Parsing.p_module(s, pxd, full_module_name)
building 'gpu_nms' extension
/usr/local/cuda/bin/nvcc -I/home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/venv/lib/python3.5/site-packages/numpy/core/include -I/usr/local/cuda/include -I/home/juggernaut/Desktop/DeepHighResolution/deep-high-resolution-net.pytorch/venv/include -I/usr/include/python3.5m -c nms_kernel.cu -o build/temp.linux-x86_64-3.5/nms_kernel.o -arch=sm_35 --ptxas-options=-v -c --compiler-options '-fPIC'
gcc: error: nms_kernel.cu: No such file or directory
gcc: warning: ‘-x c++’ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
error: command '/usr/local/cuda/bin/nvcc' failed with exit status 1

Seems like a file is missing.

Anyway, if its relevant, all packages installed -

Package         Version    
--------------- -----------
cloudpickle     0.8.0      
cycler          0.10.0     
Cython          0.29.5     
dask            1.1.2      
decorator       4.3.2      
easydict        1.7        
json-tricks     3.12.4     
kiwisolver      1.0.1      
matplotlib      3.0.2      
networkx        2.2        
numpy           1.16.1     
opencv-python   3.4.1.15   
pandas          0.24.1     
Pillow          5.4.1      
pip             10.0.1     
protobuf        3.6.1      
pyparsing       2.3.1      
python-dateutil 2.8.0      
pytz            2018.9     
PyWavelets      1.0.2      
PyYAML          3.13       
scikit-image    0.14.2     
scipy           1.2.1      
setuptools      39.1.0     
Shapely         1.6.4      
six             1.12.0     
tensorboardX    1.6        
toolz           0.9.0      
torch           1.0.1.post2
torchvision     0.2.1      
yacs            0.1.5

very low accuracy test on one GPU

I don't have four GPUs.If I run the command that you provide:

python tools/test.py \
    --cfg experiments/coco/hrnet/w32_256x192_adam_lr1e-3.yaml \
    TEST.MODEL_FILE models/pytorch/pose_coco/pose_hrnet_w32_256x192.pth \
    TEST.USE_GT_BBOX False

i got an error about device_ids.
So I change the code in test.py:
model = torch.nn.DataParallel(model,device_ids=cfg.GPUS).cuda() to
model = torch.nn.DataParallel(model, device_ids=(0,)).cuda()

the other:
num_workers=cfg.WORKERS to
num_workers=0,

It works but the model 'pose_hrnet_w32_256x192.pth' gets very low accuracy(nearly 0).I think somewhere I changed wrong or some code still need to be changed.Could you help me?

How to use our own datasets to get keypoint of person ??

Now, i want to use your trained network to get my keypoint person in our dataset .but i don't know what parts we should change?

Inference Frame Rate

@leoxiaobin Thanks for the this good work. I am wondering how much faster is the inference in term of fps? Do you have a script to run the inference model with a webcam?

How to get center location and scale when inference?

I need to get the final result which seems can calling this functions:

def get_final_preds(config, batch_heatmaps, center, scale):

But it needs a center and a scale. How to get it anyway?

AssertionError: Invalid device id

Have you tried using intermediate supervision?

Thanks for your works! Have you tried using intermediate supervision? Do you think it would be helpful to include intermediate supervision in the model?

about USE_DIFFERENT_JOINTS_WEIGHT

Hi @leoxiaobin @sunke123 @welleast
Thanks for sharing your work!

I am curious about the parameter:
LOSS.USE_DIFFERENT_JOINTS_WEIGHT

    if self.use_different_joints_weight:
        target_weight = np.multiply(target_weight, self.joints_weight)

What does it mean and when to use it?

Meet problem when download for coco data

there is no file of annotations ,what's the matter?

ModuleNotFoundError: No module named 'nms.cpu_nms'

myubuntu@myubuntu-Precision-WorkStation-T5500:~/Desktop/deep-high-resolution-net.pytorch-master$ python tools/test.py \ --cfg experiments/coco/hrnet/w32_256x192_adam_lr1e-3.yaml \ TEST.MODEL_FILE models/pytorch/pose_coco/pose_hrnet_w32_256x192.pth \ TEST.USE_GT_BBOX False
Traceback (most recent call last):
File "tools/test.py", line 31, in
import dataset
File "/home/myubuntu/Desktop/deep-high-resolution-net.pytorch-master/tools/../lib/dataset/init.py", line 12, in
from .coco import COCODataset as coco
File "/home/myubuntu/Desktop/deep-high-resolution-net.pytorch-master/tools/../lib/dataset/coco.py", line 22, in
from nms.nms import oks_nms
File "/home/myubuntu/Desktop/deep-high-resolution-net.pytorch-master/tools/../lib/nms/nms.py", line 13, in
from .cpu_nms import cpu_nms
ModuleNotFoundError: No module named 'nms.cpu_nms'

Error about 'mean is not a valid value for reduction' in Pytorch 0.4.1

Impressive work.
When I tried to run the test code, I encounter a error saying that:
'ValueError: mean is not a valid value for reduction'

I am using Python 3.6, Pytorch 0.4.1,
In my case, if i change the line 'self.criterion = nn.MSELoss(reduction='mean')'
of lib/core/loss.py from 'mean' to 'elementwise_mean', it works. I think that is the error for the Pytorch 0.4.1 which they fixed it in the later version. Hope that helps others who had the same error.

how to set SIGMA?

how to set SIGMA?why 2?

TypeError: init() got an unexpected keyword argument 'serialized_options'

protoc --version
libprotoc 3.4.0
and protobuf versuon is 3.4.0. I would be grateful if someone could provide some help.

在JointsDataset.py文件中输入图像进行warpAffine变换后的逆过程在哪个函数中

您好！
能不能想请教您一个问题，
调用function.py文件中train()函数对应行
for i, (input, target, target_weight, meta) in enumerate(val_loader):
outputs = model(input)

跳转到JointsDataset.py文件中，
def getitem(self, idx):
....
retrun input,....
调用__getitem_之后，返回的是图像仿射变换cv2.warpAffine然后transform后的数据，大小为[3,image_size[0],image_size[0]] 。

从JointsDataset.py中的返回的input（大小为[3,image_size[0],image_size[0]]）到 function.py文件train()函数中的input（大小为[batch_size,3,image_size[0],image_size[0]]）的中间操作在哪个函数中的了？也就是想问一下您，仿射变换的逆过程是在哪个函数中，因为一直无法debug进去，能不能帮忙解答一下，非常谢谢！

Recommend Projects

React

A declarative, efficient, and flexible JavaScript library for building user interfaces.
Vue.js

🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript

TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
TensorFlow

An Open Source Machine Learning Framework for Everyone
Django

The Web framework for perfectionists with deadlines.
Laravel

A PHP framework for web artisans
D3

Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

javascript

JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
web

Some thing interesting about web. New door for the world.
server

A server is a program made to process requests and deliver data to clients.
Machine learning

Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Visualization

Some thing interesting about visualization, use data art
Game

Some thing interesting about game, make everyone happy.

Recommend Org

Facebook

We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft

Open source projects and samples from Microsoft.
Google

Google ❤️ Open Source for everyone.
Alibaba

Alibaba Open Source for everyone
D3

Data-Driven Documents codes.
Tencent

China tencent open source team.