Action Classification on Kinetics 600 (val)
90.3Top-1 AccMTV-H
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MTV-HTraining Data=in-house labeled (WTS 280M), Resolution=280x280, Views=4 x 3, TFLOPS=73.572023.03 | 90.3 | 98.5 | |
| VideoMAE V2-gResolution=64 x 266x266, Views=2 x 3, TFLOPS=160.32023.03 | 89.9 | 98.5 | |
| VideoMAE V2-gViews=5 x 3, TFLOPS=38.162023.03 | 88.8 | 98.2 | |
| VideoMAE V2-HViews=5 x 3, TFLOPS=17.882023.03 | 88.3 | 98.1 | |
| MViTv2-LResolution=352x352, Views=40 x 3, TFLOPS=45.482023.03 | 87.9 | 97.9 | |
| CoVeRTraining Data=in-house labeled (JFT-3B), Views=1 x 32023.03 | 87.9 | 97.8 | |
| MaskFeatViews=1 x 10, TFLOPS=3.772023.03 | 86.4 | 97.4 | |
| MTV-BResolution=320x320, Views=4 x 3, TFLOPS=11.162023.03 | 84 | 96.2 | |
| ViViT-L FEViews=1 x 3, TFLOPS=11.942023.03 | 82.9 | 94.6 | |
| TimeSformer-LViews=1 x 3, TFLOPS=7.142023.03 | 82.2 | 95.6 | |
| SlowFast R101-NLBackbone=ResNet-101-NL, Views=10 x 3, TFLOPS=7.022023.03 | 81.8 | 95.1 |