Towards Training Stronger Video Vision Transformers for EPIC-KITCHENS-100 Action Recognition
This work addresses action recognition in video data, specifically for kitchen activities, but it is incremental as it focuses on training refinements rather than novel architectural changes.
The paper tackled improving video vision transformers for action recognition on the EPIC-KITCHENS-100 dataset by exploring training techniques like augmentations and resolutions, resulting in a single model achieving 47.4% accuracy on the validation set, a 3.4% improvement over prior work.
With the recent surge in the research of vision transformers, they have demonstrated remarkable potential for various challenging computer vision applications, such as image recognition, point cloud classification as well as video understanding. In this paper, we present empirical results for training a stronger video vision transformer on the EPIC-KITCHENS-100 Action Recognition dataset. Specifically, we explore training techniques for video vision transformers, such as augmentations, resolutions as well as initialization, etc. With our training recipe, a single ViViT model achieves the performance of 47.4\% on the validation set of EPIC-KITCHENS-100 dataset, outperforming what is reported in the original paper by 3.4%. We found that video transformers are especially good at predicting the noun in the verb-noun action prediction task. This makes the overall action prediction accuracy of video transformers notably higher than convolutional ones. Surprisingly, even the best video transformers underperform the convolutional networks on the verb prediction. Therefore, we combine the video vision transformers and some of the convolutional video networks and present our solution to the EPIC-KITCHENS-100 Action Recognition competition.