Disentangling Static and Dynamic Information for Reducing Static Bias in Action Recognition
This addresses the issue of poor performance in real-world and zero-shot action recognition for computer vision applications, but appears incremental as it builds on existing bias reduction techniques.
The paper tackled the problem of static bias in action recognition models, where they rely too much on static cues instead of dynamic motion, and proposed a method using statistical independence and scene prediction losses to reduce this bias, showing effective reduction in experiments.
Action recognition models rely excessively on static cues rather than dynamic human motion, which is known as static bias. This bias leads to poor performance in real-world applications and zero-shot action recognition. In this paper, we propose a method to reduce static bias by separating temporal dynamic information from static scene information. Our approach uses a statistical independence loss between biased and unbiased streams, combined with a scene prediction loss. Our experiments demonstrate that this method effectively reduces static bias and confirm the importance of scene prediction loss.