Extending Temporal Data Augmentation for Video Action Recognition

Pixel space augmentation has grown in popularity many Deep Learning areas, due to its effectiveness, simplicity, and low computational cost. Data for videos, however, still remains an under-explored research topic, as most works have been treating inputs stacks of static images rather than temporally linked series data. Recently, it shown that involving the time dimension when designing augmentations can be superior spatial-only variants video action recognition [34]. In this paper, we propose several novel enhancements these techniques strengthen relationship between spatial temporal domains achieve a deeper level perturbations. The results our outperform their respective Top-1 Top-5 settings on UCF-101 [55] HMDB-51 [38] datasets.