Robust video action recognition on UCF50 under realistic corruption shifts, with reproducible corruption generation, baseline training/evaluation, ViTTA test-time adaptation, and RMGA adaptation benchmarking.
computer-vision pytorch course-project action-recognition torchvision domain-shift vitta video-corruption test-time-adaptation corruption-robustness ucf50-dataset low-compute rmga
-
Updated
Apr 25, 2026 - Jupyter Notebook