DeepPose

This implements training of DeepPose on the COCO2017 dataset, mainly modified from mmpose.

DeepPose Detail

For details, see mmpose/models/backbones/resnet.py.

Requirements

• Install PyTorch (pytorch.org) and apex

• pip install -r requirements.txt

  Caution: If your cpu is based on ARM，you need to download the source code of package xtcocoapi. You can download the source code from GoogleDrive and do like this:

  unzip xtcocoapi.zip
  cd xtcocoapi
  python3 setup.py build_ext install
  

• HRNet-Human-Pose-Estimation provides person detection result of COCO val2017 to reproduce our multi-person pose estimation results. Please download from OneDrive or GoogleDrive. Download and extract them , and place COCO_val2017_detections_AP_H_56_person.json under $DeepPose/person_detection_results.

Training

# default work directory is work_dirs/npu_deeppose_res50_coco_256x192/
# O2 training , defalut device 0
bash test/train_full_1p.sh --data_path=coco2017_data_path

# O2 training 8p
bash test/train_full_8p.sh --data_path=coco2017_data_path

# eval 8p
# default ckpt path is work_dirs/npu_deeppose_res50_coco_256x192/epoch_210.pth
# you need to choose the correct path of checkpoint
bash bash test/train_full_8p.sh --data_path=coco2017_data_path --checkpoint=ckpt_path

# online inference demo
python3 demo.py configs/top_down/deeppose/coco/npu_deeppose_res50_coco_256x192.py work_dirs/npu_deeppose_res50_coco_256x192/epoch_210.pth

# onnx
python3 pthtar2onnx.py

DeepPose training result

名称	精度	性能	AMP_Type
GPU-1p	-	194	O2
GPU-8p	52.50	1160	O2
NPU-1p	-	117	O2
NPU-8p	52.65	650-830	O2

Statement

For details about the public address of the code in this repository, you can get from the file public_address_statement.md