Code repository for the paper SMPR: Single-Stage Multi-Person Pose Regression, by Junqi Lin, Huixin Miao, Junjie Cao, Zhixun Su and Risheng Liu.
Results on COCO test-dev.
Backbone | mAP | AP^{50} | AP^{75} | AP^{M} | AP^{L} |
---|---|---|---|---|---|
ResNet50 | 62.6 | 85.9 | 68.6 | 56.1 | 71.7 |
ResNet50 (multi-testing) | 65.3 | 87.9 | 72.1 | 59.8 | 73.3 |
HRNet-w32 | 68.2 | 88.7 | 75.3 | 63.3 | 75.4 |
HRNet-w32 (multi-testing) | 70.2 | 89.7 | 77.5 | 65.9 | 77.2 |
conda create -n mmdet python=3.7
conda activate mmdet
conda install pytorch=1.4.0 cudatoolkit=10.1 torchvision=0.5.0
pip install cython
git clone https://github.com/cocodataset/cocoapi.git
cd cocoapi/PythonAPI
python setup.py build_ext --inplace
python setup.py build_ext install
pip install -r requirements.txt
pip install Pillow==6.2.2
pip install -v -e .
put train2017 and val2017 in data/coco
put person_keypoints_train2017.json and person_keypoints_val2017.json in data/coco/annotations
cd data/coco/annotations
% generate 'person_keypoints_train2017_pesudobox.json'
python pesudo_box_train.py
% generate 'person_keypoints_val2017_pesudobox.json'
python pesudo_box_val.py
You can download the trained model on Baidu Yun,with the extraction code:aaaa
You can now evaluate the models on the COCO val2017 split:
./tools/dist_test.sh configs/SMPR/ResNet_50.py work_dirs/r50.pth 4 --eval keypoints --options "jsonfile_prefix=./work_dirs/r50"
@misc{SMPR2020,
Author = {Junqi Lin and Huixin Miao and Junjie Cao and Zhixun Su and Risheng Liu},
Title = {SMPR: Single-Stage Multi-Person Pose Regression},
Year = {2020},
Eprint = {arXiv:2006.15576},
}
We would like to thank MMDetection team for producing this great object detection toolbox!
This project is released under the Apache 2.0 license.