AI-challenger-Realtime_Multi-Person_Pose_Estimation

Training

1. Download the original project Realtime multi-person and add the files in my repository and put them in the trainingdirectory.
2. Run cd training; Download the AI-challenger dataset and put the images in dataset/COCO/images/, keypoints annotations in dataset/COCO/annotations/ and COCO official toolbox in dataset/COCO/coco/.

dataset/AI_challenge/train/
--------------------/train/keypoint_train_images_20170902/
--------------------/train/keypoint_train_annotations_20170909.json

3. Python ai2coco_art_json.pyto transform the annotation format of the AI-challenger dataset to the MS COCO annotation format. The final annotation file is coco_artkeypoint_train_annotations_20170909.json
4. Run my_getANNO.m in matlab to convert the annotation format from json to mat in dataset/AI_challenge/mat/
5. Run my_genCOCOMask.m in matlab to obatin the mask images for unlabeled person. You can use 'parfor' in matlab to speed up the code.(I use the human bbox as the segmentation annotations)
6. Run python my_genLMDB.py to generate your LMDB.
7. Download our modified caffe: caffe_train. Compile pycaffe. It will be merged with caffe_rtpose (for testing) soon.
8. Run python setLayers.py --exp 1 to generate the prototxt and shell file for training.
9. Download VGG-19 model, we use it to initialize the first 10 layers for training.
10. Run bash train_pose.sh 0,1 (generated by setLayers.py) to start the training with two gpus.

Citation

Please cite the paper in your publications if it helps your research:

@inproceedings{cao2017realtime,
  author = {Zhe Cao and Tomas Simon and Shih-En Wei and Yaser Sheikh},
  booktitle = {CVPR},
  title = {Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields},
  year = {2017}
  }
            
@inproceedings{wei2016cpm,
  author = {Shih-En Wei and Varun Ramakrishna and Takeo Kanade and Yaser Sheikh},
  booktitle = {CVPR},
  title = {Convolutional pose machines},
  year = {2016}
  }