Rafal Pytel
Paper download is intended for registered attendees only, and is
subjected to the IEEE Copyright Policy. Any other use is strongly forbidden.
Papers from this author
Tilting at Windmills: Data Augmentation for Deeppose Estimation Does Not Help with Occlusions
Rafal Pytel, Osman Semih Kayhan, Jan Van Gemert
Auto-TLDR; Targeted Keypoint and Body Part Occlusion Attacks for Human Pose Estimation
Abstract Slides Poster Similar
Occlusion degrades the performance of human poseestimation. In this paper, we introduce targeted keypoint andbody part occlusion attacks. The effects of the attacks are system-atically analyzed on the best performing methods. In addition, wepropose occlusion specific data augmentation techniques againstkeypoint and part attacks. Our extensive experiments show thathuman pose estimation methods are not robust to occlusion anddata augmentation does not solve the occlusion problems.