Tilting at windmills: Data augmentation for deep pose estimation does not help with occlusions

Tilting at windmills: Data augmentation for deep pose estimation does not help with occlusions

Occlusion degrades the performance of human pose estimation. In this paper, we introduce targeted keypoint and body part occlusion attacks. The effects of the attacks are systematically analyzed on the best performing methods. In addition, we propose occlusion specific data augmentation techniques a...

Full description

Saved in:
Bibliographic Details
Published in:2020 25th International Conference on Pattern Recognition (ICPR) pp. 10568 - 10575
Main Authors: Pytel, Rafal, Kayhan, Osman Semih, van Gemert, Jan C.
Format: Conference Proceeding
Language:English
Published: IEEE 10-01-2021
Subjects:
Pattern recognition
Pose estimation
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Occlusion degrades the performance of human pose estimation. In this paper, we introduce targeted keypoint and body part occlusion attacks. The effects of the attacks are systematically analyzed on the best performing methods. In addition, we propose occlusion specific data augmentation techniques against keypoint and part attacks. Our extensive experiments show that human pose estimation methods are not robust to occlusion and data augmentation does not solve the occlusion problems. 1 1 For the code and the extended version: https://github.com/rpytell/ocdusion-vs-data-augmentations
DOI:10.1109/ICPR48806.2021.9412475