Towards Training Stronger Video Vision Transformers for EPIC-KITCHENS-100 Action Recognition
Abstract
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.
Keywords
Cite
@article{arxiv.2106.05058,
title = {Towards Training Stronger Video Vision Transformers for EPIC-KITCHENS-100 Action Recognition},
author = {Ziyuan Huang and Zhiwu Qing and Xiang Wang and Yutong Feng and Shiwei Zhang and Jianwen Jiang and Zhurong Xia and Mingqian Tang and Nong Sang and Marcelo H. Ang},
journal= {arXiv preprint arXiv:2106.05058},
year = {2021}
}
Comments
CVPRW 2021, EPIC-KITCHENS-100 Competition Report