YOLOF: You Only Look At One-level Feature

This is a PyTorch version of YOLOF.

Requirements

We recommend you to use Anaconda to create a conda environment:

conda create -n yolof python=3.6

Then, activate the environment:

conda activate yolof

Requirements:

pip install -r requirements.txt

We suggest that PyTorch should be higher than 1.9.0 and Torchvision should be higher than 0.10.3. At least, please make sure your torch is version 1.x.

My environment:

Torch = 1.9.1
Torchvision = 0.10.1

Main results on COCO-val

AP results of YOLOF

Model	scale	FPS	AP	AP50	Weight
YOLOF-R18-C5_1x	800,1333	100	32.6	51.3	github
YOLOF-R50-C5_1x	800,1333	50	37.5	57.4	github
YOLOF-R50-DC5_1x	800,1333	32	38.7	58.5	github
YOLOF-R101-C5_1x	800,1333				github
YOLOF-R101-DC5_1x	800,1333				github
YOLOF-RT-R50_3x	512,736	60	39.4	58.6	github

Limited by my computing resources, I cannot train YOLOF_R101_C5_1x, YOLOF_R101_DC5_1x. I would be very grateful if you used this project to train them and would like to share weight files.

Visualization

(YOLOF_R50-C5_1x)

AP results of FCOS

Model	scale	FPS	AP	AP50	Weight
FCOS-R18_1x	800,1333	42	33.0	51.3	github
FCOS-R50_1x	800,1333	30	38.2	58.0	github
FCOS-RT-R18_4x	512,736	83	33.8	51.5	github
FCOS-RT-R50_4x	512,736	60	38.7	58.0	github
FCOS-RT-R18-OTA_4x	512,736
FCOS-RT-R50-OTA_4x	512,736

FCOS-RT-R18-OTA_4x means that we use the SimOTA to train the FCOS-RT-R18.

AP results of RetiniaNet

Model	scale	AP	AP50	Weight
RetinaNet-R18_1x	800,1333	30.8	49.6	github
RetinaNet-R50_1x	800,1333
RetinaNet-RT-R18_4x	512,736
RetinaNet-RT-R50_4x	512,736

Train

Single GPU

You can run the following command:

python train.py \
        --cuda \
        -d coco \
        --root /mnt/share/ssd2/dataset/ \
        -v yolof-r50 \
        --batch_size 16 \
        --schedule 1x \
        --grad_clip_norm 4.0 \

or, you just run the script:

sh train.sh

You can change the configurations of train.sh, according to your own situation.

Multi GPUs

You can run the following command:

# 2 GPUs
python -m torch.distributed.run --nproc_per_node=2 train.py \
                                                    --cuda \
                                                    -dist \
                                                    -d coco \
                                                    --root /mnt/share/ssd2/dataset/ \
                                                    -v yolof-r50 \
                                                    --batch_size 8 \
                                                    --grad_clip_norm 4.0 \
                                                    --num_workers 4 \
                                                    --schedule 1x \

or, you just run the script:

sh train_ddp.sh

Test

python test.py -d coco \
               --cuda \
               -v yolof-r50 \
               --weight path/to/weight \
               --root path/to/dataset/ \
               --show

Evaluate

python eval.py -d coco-val \
               --cuda \
               -v yolof-r50 \
               --weight path/to/weight \
               --root path/to/dataset/ \

Our AP results of YOLOF-R50-C5-1x on COCO-val:

Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.375
Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.574
Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.399
Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.185
Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.421
Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.522
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.312
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.507
Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.552
Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.332
Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.625
Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.739

Demo

I have provide some images in data/demo/images/, so you can run following command to run a demo:

python demo.py --mode image \
               --path_to_img data/demo/images/ \
               -v yolof-r50 \
               --cuda \
               --weight path/to/weight \
               --show

If you want run a demo of streaming video detection, you need to set --mode to video, and give the path to video --path_to_vid。

python demo.py --mode video \
               --path_to_img data/demo/videos/your_video \
               -v yolof-r50 \
               --cuda \
               --weight path/to/weight \
               --show

If you want run video detection with your camera, you need to set --mode to camera。

python demo.py --mode camera \
               -v yolof-r50 \
               --cuda \
               --weight path/to/weight \
               --show

yjh0410 / pytorch_yolof Goto Github PK

pytorch_yolof's Introduction

YOLOF: You Only Look At One-level Feature

Requirements

Main results on COCO-val

Train

Test

Evaluate

Demo

pytorch_yolof's People

Contributors

Stargazers

Watchers

Forkers

pytorch_yolof's Issues

Recommend Projects

Recommend Topics

Recommend Org