Cascade-keypoint-localization
An end-to-end cascade keypoint detection model implemented using Pytorch. Also used in FashionAI challenge.
We design a two-stage model to predict the location of keypoints.
- stage 1: predict a rough location
.
- stage 2: use the location predicted in the first stage as the center and crop an area of 31*31 to extract local features of the keypoint. These local features will be used to precdict the differece between the prediction of stage 1 and the groundtrouth
.
Example:
