The original code is from "https://github.com/matterport/Mask_RCNN" on Python 3, Keras, and TensorFlow. The code reproduce the work of "https://arxiv.org/abs/1703.06870" but does not include code for keypoints. So the work that ralice in this publication is: # modify the code to perform the detection of keypoints.
The trainig file is "main", you can biging with thid file.
The dataset is in https://challenger.ai/competition/keypoint/subject
This image is one of the first tests, there is no mean average presition mAP The resunts are far from the original paper
This picture reprecents the last layer in the network
P.S: modified json files are uploaded in release 2.