DensePose-Paddle

A reproduction of DensePose by PaddlePaddle

1. 首先复现detectron2

并没有使用facebook的代码，而是https://github.com/MILVLG/bottom-up-attention.pytorch

2. 训练

首先准备数据集 dataset 格式如下:

datasets/coco/
  annotations/
    densepose_{train,minival,valminusminival}2014.json
    densepose_minival2014_100.json   (optional, for testing only)
  {train,val}2014/
    # image files that are mentioned in the corresponding json

训练过程先略去

3. 测试图片

（运行代码和facebook的detectron2_0.6版本有些不同） （1） Show bounding box and segmentation:

python apply_net.py show configs/densepose_rcnn_R_50_FPN_s1x.yaml model_final_162be9.pkl 1.jpg dp_segm,bbox --output 1_segm.png

2. Show bounding box and estimated U coordinates for body parts:

python apply_net.py show configs/densepose_rcnn_R_50_FPN_s1x.yaml model_final_162be9.pkl 1.jpg dp_u,bbox --output 1_u.png

3. Show bounding box and estimated V coordinates for body parts:

python apply_net.py show configs/densepose_rcnn_R_50_FPN_s1x.yaml model_final_162be9.pkl 1.jpg dp_v,bbox --output 1_v.png

4. Show bounding box and estimated U and V coordinates via contour plots:

python apply_net.py show configs/densepose_rcnn_R_50_FPN_s1x.yaml model_final_162be9.pkl 1.jpg dp_contour,bbox --output 1_contour.png