Joint human pose and part segmentation

Course: Visual Recognition, Indian Institute of Technology Kanpur, 2017

Developed a system to jointly predict human pose and part segments using an end-to-end trainable model using data containing partial ground truth

  • Developed a new architecture by combining CVPR 2017 paper “Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields” with fcn architecture for segmentation
  • Trained the model using relevant parts from MPII and PASCAL VOC part datasets, with the implementation in PyTorch.

Report Poster