Domain Generalization through Audio-Visual Relative Norm Alignment in First Person Action Recognition
This addresses cross-domain issues in egocentric activity recognition, enabling better generalization to unseen scenarios for applications with wearable cameras.
The paper tackles the problem of domain generalization in first-person action recognition by introducing a Relative Norm Alignment loss to re-balance audio-visual modalities, achieving strong results on EPIC-Kitchens datasets.
First person action recognition is becoming an increasingly researched area thanks to the rising popularity of wearable cameras. This is bringing to light cross-domain issues that are yet to be addressed in this context. Indeed, the information extracted from learned representations suffers from an intrinsic "environmental bias". This strongly affects the ability to generalize to unseen scenarios, limiting the application of current methods to real settings where labeled data are not available during training. In this work, we introduce the first domain generalization approach for egocentric activity recognition, by proposing a new audio-visual loss, called Relative Norm Alignment loss. It re-balances the contributions from the two modalities during training, over different domains, by aligning their feature norm representations. Our approach leads to strong results in domain generalization on both EPIC-Kitchens-55 and EPIC-Kitchens-100, as demonstrated by extensive experiments, and can be extended to work also on domain adaptation settings with competitive results.