Action Classification on Kinetics-400 (test)
87.2AccuracyVideoMAE V2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VideoMAE V2Backbone=ViT-g, pre-train data=UnlabeledHybrid, data size=1.35M, epoch=1200, Input Resolution=16x224x224, Sampling Stride=42023.03 | 87.2 | 83.9 | |
| VideoMAE V2Backbone=ViT-H, pre-train data=UnlabeledHybrid, data size=1.35M, epoch=1200, Input Resolution=16x224x224, Sampling Stride=42023.03 | 86.9 | 83.2 | |
| VideoMAE V1Backbone=ViT-H, pre-train data=Kinetics400, data size=0.24M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 86.6 | — | |
| VideoMAE V2Backbone=ViT-L, pre-train data=UnlabeledHybrid, data size=1.35M, epoch=1200, Input Resolution=16x224x224, Sampling Stride=42023.03 | 85.4 | 81.3 | |
| VideoMAE V1Backbone=ViT-L, pre-train data=Kinetics400, data size=0.24M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 85.2 | — | |
| MAE-STBackbone=ViT-H, pre-train data=Kinetics400, data size=0.24M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 85.1 | — | |
| MAE-STBackbone=ViT-L, pre-train data=Kinetics400, data size=0.24M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 84.8 | — | |
| MAE-STBackbone=ViT-L, pre-train data=IG-uncurated, data size=1M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 84.4 | — | |
| VideoMAE V1Backbone=ViT-B, pre-train data=Kinetics400, data size=0.24M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 81.5 | — | |
| VideoMAE V2Backbone=ViT-B, pre-train data=UnlabeledHybrid, data size=1.35M, epoch=1200, Input Resolution=16x224x224, Sampling Stride=42023.03 | 81.5 | 77 | |
| MAE-STBackbone=ViT-B, pre-train data=Kinetics400, data size=0.24M, epoch=1600, Input Resolution=16x224x224, Sampling Stride=42023.03 | 81.3 | — | |
| VFCZero-shot=true2023.04 | 58.8 | — | |
| BaselineZero-shot=true2023.04 | 55.6 | — |