whole_body_detection

dataset:

based on coco: https://cocodataset.org/#home and coco-whole-body: https://github.com/jin-s13/COCO-WholeBody
add keypoints from 17(coco) to 35(6 for each hand, 3 for each foot, total: 18, from coco-whole-body)
more detail see load_ann.py

model:

detection part: using faster rcnn and load checkpoints from pretrained model.
pose part: using resNet_50 and change input size as 256x256, output size as 64x64. Trained from pretrained model for 17 coco model.
more detail see train_model.py