Akash Sengupta, Ignas Budvytis, Roberto Cipolla
ICCV 2021
[paper+supplementary][poster][results video]

This is the official code repository of the above paper, which takes a probabilistic approach to 3D human shape and pose estimation and predicts multiple plausible 3D reconstruction samples given an input image.

This repository contains inference, training and evaluation code. A few weaknesses of this approach, and future research directions, are listed below. If you find this code useful in your research, please cite the following publication:

@InProceedings{sengupta2021hierprobhuman,
               author = {Sengupta, Akash and Budvytis, Ignas and Cipolla, Roberto},
               title = {{Hierarchical Kinematic Probability Distributions for 3D Human Shape and Pose Estimation from Images in the Wild}},
               booktitle = {International Conference on Computer Vision},
               month = {October},
               year = {2021}                         
}

• Linux or macOS
• Python ≥ 3.6

We recommend using a virtual environment to install relevant dependencies:

python3 -m venv HierProbHuman
source HierProbHuman/bin/activate

Install torch and torchvision (the code has been tested with v1.6.0 of torch), as well as other dependencies:

pip install torch==1.6.0 torchvision==0.7.0
pip install -r requirements.txt

Finally, install pytorch3d, which we use for data generation during training and visualisation during inference. To do so, you will need to first install the CUB library following the instructions here. Then you may install pytorch3d - note that the code has been tested with v0.3.0 of pytorch3d, and we recommend installing this version using:

pip install "git+https://github.com/facebookresearch/[email protected]"

IMPORTANT: if you would like to use pytorch3d v0.5.0 or greater, you will need to slightly modify the camera models in the renderer class definition, as outlined here.

You will need to download the SMPL model. The neutral model is required for training and running the demo code. If you want to evaluate the model on datasets with gendered SMPL labels (such as 3DPW and SSP-3D), the male and female models are available here. You will need to convert the SMPL model files to be compatible with python3 by removing any chumpy objects. To do so, please follow the instructions here.

Download pre-trained model checkpoints for our 3D Shape/Pose network, as well as for 2D Pose HRNet-W48 from here. In addition to the neutral-gender prediction network presented in the paper, we provide pre-trained checkpoints for male and female prediction networks, which are trained with male/female SMPL shape respectively. Download these checkpoints if you wish to do gendered shape inference.

Place the SMPL model files and network checkpoints in the model_files directory, which should have the following structure. If the files are placed elsewhere, you will need to update configs/paths.py accordingly.

HierarchicalProbabilistic3DHuman
├── model_files                                  # Folder with model files
│   ├── smpl
│   │   ├── SMPL_NEUTRAL.pkl                     # Gender-neutral SMPL model
│   │   ├── SMPL_MALE.pkl                        # Male SMPL model
│   │   ├── SMPL_FEMALE.pkl                      # Female SMPL model
│   ├── poseMF_shapeGaussian_net_weights.tar     # Pose/Shape distribution predictor checkpoint
│   ├── pose_hrnet_w48_384x288.pth               # Pose2D HRNet checkpoint
│   ├── cocoplus_regressor.npy                   # Cocoplus joints regressor
│   ├── J_regressor_h36m.npy                     # Human3.6M joints regressor
│   ├── J_regressor_extra.npy                    # Extra joints regressor
│   └── UV_Processed.mat                         # DensePose UV coordinates for SMPL mesh             
└── ...

run_predict.py is used to run inference on a given folder of input images. For example, to run inference on the demo folder, do:

python run_predict.py --image_dir ./demo/ --save_dir ./output/ --visualise_samples --visualise_uncropped

This will first detect human bounding boxes in the input images using Mask-RCNN. If your input images are already cropped and centred around the subject of interest, you may skip this step using --cropped_images as an option. The 3D Shape/Pose network is somewhat sensitive to cropping and centering - this is a good place to start troubleshooting in case of poor results.

If the gender of the subject is known, you may wish to carry out gendered inference using the provided male/female model weights. This can be done by modifying the above command as follows:

python run_predict.py --gender male --pose_shape_weights model_files/poseMF_shapeGaussian_net_weights_male.tar --image_dir ./demo/ --save_dir ./output_male/ --visualise_samples --visualise_uncropped

(similar for the female model). Using gendered models for inference may result in better body shape estimates, as it serves as a prior over 3D shape.

Inference can be slow due to the rejection sampling procedure used to estimate per-vertex 3D uncertainty. If you are not interested in per-vertex uncertainty, you may modify predict/predict_poseMF_shapeGaussian_net.py by commenting out code related to sampling, and use a plain texture to render meshes for visualisation (this will be cleaned up and added as an option to in the run_predict.py future).

run_evaluate.py is used to evaluate our method on the 3DPW and SSP-3D datasets. A description of the metrics used to measure performance is given in metrics/eval_metrics_tracker.py.

Download SSP-3D from here. Update configs/paths.py with the path pointing to the un-zipped SSP-3D directory. Evaluate on SSP-3D with:

python run_evaluate.py -D ssp3d

Download 3DPW from here. You will need to preprocess the dataset first, to extract centred+cropped images and SMPL labels (adapted from SPIN):

python data/pw3d_preprocess.py --dataset_path $3DPW_DIR_PATH

This should create a subdirectory with preprocessed files, such that the 3DPW directory has the following structure:

$3DPW_DIR_PATH
      ├── test                                  
      │   ├── 3dpw_test.npz    
      │   ├── cropped_frames   
      ├── imageFiles
      └── sequenceFiles

Additionally, download HRNet 2D joint detections on 3DPW from here, and place this in $3DPW_DIR_PATH/test. Update configs/paths.py with the path pointing to $3DPW_DIR_PATH/test. Evaluate on 3DPW with:

python run_evaluate.py -D 3dpw

The number of samples used to evaluate sample-related metrics can be changed using the --num_samples option (default is 10).

run_train.py is used to train our method using random synthetic training data (rendered on-the-fly during training).

Download .npz files containing SMPL training/validation body poses and textures from here. Place these files in a ./train_files directory, or update the appropriate variables in configs/paths.py with paths pointing to the these files. Note that the SMPL textures are from SURREAL and MultiGarmentNet.

We use images from LSUN as random backgrounds for our synthetic training data. Specifically, images from the 10 scene categories are used. Instructions to download and extract these images are provided here. The copy_lsun_images_to_train_files_dir.py script can be used to copy LSUN background images to the ./train_files directory, which should have the following structure:

train_files
      ├── lsun_backgrounds
          ├── train
          ├── val
      ├── smpl_train_poses.npz
      ├── smpl_train_textures.npz                                  
      ├── smpl_val_poses.npz                                  
      └── smpl_val_textures.npz                                  

Finally, start training with:

python run_train.py -E experiments/exp_001

As a sanity check, the script should find 91106 training poses, 125 + 792 training textures, 397582 training backgrounds, 33347 validation poses, 32 + 76 validation textures and 3000 validation backgrounds.

The following aspects of our method may be the subject of future research:

• Mesh interpenetrations: this occurs occasionally amongst 3D mesh samples drawn from shape and pose distribution predictions. A sample inter-penetratation penalty may be useful.
• Sample diversity / distribution expressiviness: since the predicted distributions are uni-modal, sample diversity may be limited.
• Sampling speed: rejection sampling from a matrix-Fisher distribution is currently slow.
• Non-tight clothing: body shape prediction accuracy suffers when subjects are wearing non-tight clothing, since the synthetic training data does not model clothing in 3D (only uses clothing textures). Perhaps better synthetic data (e.g. AGORA) will alleviate this issue.

Code was adapted from/influenced by the following repos - thanks to the authors!

• HMR
• SPIN
• VIBE
• HRNet
• PyTorch3D
• Probabilistic Orientation Estimation with Matrix Fisher Distributions
• CannyEdgePytorch
• Matrix-Fisher-Distribution
• SURREAL
• MultiGarmnetNet
• LSUN