Papers › VidTr: Video Transformer Without Convolutions

VidTr: Video Transformer Without Convolutions

23 Apr 2021ICCV 2021 10arXiv:2104.11746archive 2025-07-28

Yanyi Zhang, Xinyu Li, Chunhui Liu, Bing Shuai, Yi Zhu, Biagio Brattoli, Hao Chen, Ivan Marsic, Joseph Tighe

We introduce Video Transformer (VidTr) with separable-attention for video classification. Comparing with commonly used 3D networks, VidTr is able to aggregate spatio-temporal information via stacked attentions and provide better performance with higher efficiency. We first introduce the vanilla video transformer and show that transformer module is able to perform spatio-temporal modeling from raw pixels, but with heavy memory usage. We then present VidTr which reduces the memory cost by 3.3× while keeping the same performance. To further optimize the model, we propose the standard deviation based topK pooling for attention (pool_(topKₛtd)), which reduces the computation by dropping non-informative features along temporal dimension. VidTr achieves state-of-the-art performance on five commonly used datasets with lower computational requirement, showing both the efficiency and effectiveness of our design. Finally, error analysis and visualization show that VidTr is especially good at predicting actions that require long-term temporal reasoning.

PaperPDFConference PDF

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Action ClassificationAction RecognitionVideo Classification

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Action Classification Charades En-VidTr-L MAP 47.3 #15 of 49 Archive leaderboard report
Action Classification Charades VidTr-L MAP 43.5 #22 of 49 Archive leaderboard report
Action Classification Kinetics-400 En-VidTr-L Acc@1 80.5 #96 of 207 Archive leaderboard report
Action Classification Kinetics-400 En-VidTr-L Acc@5 94.6 #96 of 207 Archive leaderboard report
Action Classification Kinetics-400 En-VidTr-M Acc@1 79.7 #109 of 207 Archive leaderboard report
Action Classification Kinetics-400 En-VidTr-M Acc@5 94.2 #109 of 207 Archive leaderboard report
Action Classification Kinetics-400 En-VidTr-S Acc@1 79.4 #112 of 207 Archive leaderboard report
Action Classification Kinetics-400 En-VidTr-S Acc@5 94 #112 of 207 Archive leaderboard report
Action Classification Kinetics-700 En-VidTr-L Top-1 Accuracy 70.8 #22 of 36 Archive leaderboard report
Action Classification Kinetics-700 En-VidTr-L Top-5 Accuracy 89.4 #22 of 36 Archive leaderboard report
Action Classification Kinetics-700 VidTr-L Top-1 Accuracy 70.2 #24 of 36 Archive leaderboard report
Action Classification Kinetics-700 VidTr-L Top-5 Accuracy 89 #24 of 36 Archive leaderboard report
Action Classification Kinetics-700 VidTr-M Top-1 Accuracy 69.5 #25 of 36 Archive leaderboard report
Action Classification Kinetics-700 VidTr-M Top-5 Accuracy 88.3 #25 of 36 Archive leaderboard report
Action Classification Kinetics-700 VidTr-S Top-1 Accuracy 67.3 #27 of 36 Archive leaderboard report
Action Classification Kinetics-700 VidTr-S Top-5 Accuracy 87.7 #27 of 36 Archive leaderboard report
Action Recognition HMDB-51 VidTr-L Average accuracy of 3 splits 74.4 #42 of 77 Archive leaderboard report
Action Recognition Something-Something V2 VidTr-L Top-1 Accuracy 60.2 #112 of 123 Archive leaderboard report
Action Recognition UCF101 VidTr-L 3-fold Accuracy 96.7 #34 of 91 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Absolute Position EncodingsAdamAttentionBPEDense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionSoftmaxTransformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections