VidTr: Video Transformer Without Convolutions

Benchmark Model Rank Results
action-classification-on-charadesEn-VidTr-L–MAP: 47.3
action-classification-on-charadesVidTr-L–MAP: 43.5
action-classification-on-kinetics-400En-VidTr-L–Acc@1: 80.5Acc@5: 94.6
action-classification-on-kinetics-400En-VidTr-M–Acc@1: 79.7Acc@5: 94.2
action-classification-on-kinetics-400En-VidTr-S–Acc@1: 79.4Acc@5: 94
action-classification-on-kinetics-700En-VidTr-L–Top-1 Accuracy: 70.8Top-5 Accuracy: 89.4
action-classification-on-kinetics-700VidTr-L–Top-1 Accuracy: 70.2Top-5 Accuracy: 89
action-classification-on-kinetics-700VidTr-M–Top-1 Accuracy: 69.5Top-5 Accuracy: 88.3
action-classification-on-kinetics-700VidTr-S–Top-1 Accuracy: 67.3Top-5 Accuracy: 87.7
action-recognition-in-videos-on-hmdb-51VidTr-L–Average accuracy of 3 splits: 74.4
action-recognition-in-videos-on-something-something-v2VidTr-L–Top-1 Accuracy: 60.2
action-recognition-in-videos-on-ucf101VidTr-L–3-fold Accuracy: 96.7