Why Can't I Dance in the Mall? Mitigating Scene Bias in Action Recognition

Learn about scene bias in action recognition and explore techniques to mitigate its effects in this 22-minute conference talk from the University of Central Florida. Delve into the motivation behind addressing representation bias, understand its definition and measurement, and examine real-world examples. Discover proposed solutions, including adversary training and artificial occlusion, and explore related works in the field. Gain insights into testing methods, various data sets used for analysis, and quantitative results in action recognition. Examine scene representation bias comparisons and class activation maps before reaching the conclusion on mitigating scene bias in action recognition systems.

Motivation
Representation Bias
Bias Definition
Bias Measure
Bias Example
Proposed Solution
Related Works
Adversary Training
Artificial Occlusion
Crossentropy
Adversary Loss
Human Mask Confusion Loss
Testing Method
Data Sets
Human Motion Database
Diving Motion Database
Temporal Action Localization
Oshin Database
Quantitative Results
Action Recognition
Scene Representation Bias Comparison
Class Activation Maps
Conclusion