Video Action Recognition on HMDB51 (avg over all splits)
83.39Top-1 AccOur (MViTv2-S based)
Evaluation Results
| Method | Links | |
|---|---|---|
| Our (MViTv2-S based)Pre-trained=Kinetics-4002026.01 | 83.39 | |
| Two-Stream I3D [11]Pre-trained=Imagenet+Kinetics 4002026.01 | 80.7 | |
| BraVePretrain=AS, Modality=VFA, Protocol=Fine-tuned2022.11 | 79.4 | |
| XKDPretrain=AS, Modality=VA, Protocol=Fine-tuned2022.11 | 75.7 | |
| OFF [14]2026.01 | 74.2 | |
| MViTv2-S finetune (our baseline) [5]Pre-trained=Kinetics-4002026.01 | 73.59 | |
| VideoMAEPretrain=K400, Modality=V, Protocol=Fine-tuned2022.11 | 73.3 | |
| BraVePretrain=AS, Modality=VA, Protocol=Fine-tuned2022.11 | 70.8 | |
| ST-ResNet + IDT [33]2026.01 | 70.3 | |
| MMVPretrain=AS, Modality=VA, Protocol=Fine-tuned2022.11 | 70.1 | |
| BraVePretrain=AS, Modality=VFA, Protocol=Linear2022.11 | 69.9 | |
| Three-Stream TSN [20]2026.01 | 69.4 | |
| Two-Stream CNNs [12]Pre-trained=ImageNet2026.01 | 69.2 | |
| XKDPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 69 | |
| Two-Stream TSN [20]2026.01 | 68.5 | |
| CrissCrossPretrain=AS, Modality=VA, Protocol=Fine-tuned2022.11 | 67.4 | |
| ELOPretrain=YT, Modality=VFA, Protocol=Fine-tuned2022.11 | 67.4 | |
| CM-ACCPretrain=AS, Modality=VA, Protocol=Linear2022.11 | 67.2 | |
| STICAPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 67 | |
| MViTv1-B finetune (our baseline)[4]Pre-trained=Kinetics-4002026.01 | 66.75 | |
| Two-Stream I3D [11]2026.01 | 66.4 | |
| XKD-MASPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 65.9 | |
| TDD+iDT [31]2026.01 | 65.9 | |
| XKD-MATSPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 65.7 | |
| VATTPretrain=AS+HT, Modality=VAT, Protocol=Linear2022.11 | 65.2 | |
| LTC+iDT [32]2026.01 | 64.8 | |
| CrissCrossPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 64.7 | |
| AVIDPretrain=AS, Modality=VA, Protocol=Fine-tuned2022.11 | 64.7 | |
| XDCPretrain=AS, Modality=VA, Protocol=Fine-tuned2022.11 | 63.7 | |
| BraVePretrain=AS, Modality=VA, Protocol=Linear2022.11 | 63.6 | |
| CPDPretrain=K400, Modality=VT, Protocol=Linear2022.11 | 63.6 | |
| VATT-MAPretrain=AS+HT, Modality=VAT, Protocol=Linear2022.11 | 63.1 | |
| VideoMAEPretrain=U101/H51, Modality=V, Protocol=Fine-tuned2022.11 | 62.6 | |
| GDTPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 62.3 | |
| XKDPretrain=AS, Modality=VA, Protocol=Linear2022.11 | 62.2 | |
| CMACCPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 61.8 | |
| CMACPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 61.1 | |
| MIL-NCEPretrain=HT, Modality=VT, Protocol=Fine-tuned2022.11 | 61 | |
| AVIDPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 60.8 | |
| MMVPretrain=AS, Modality=VA, Protocol=Linear2022.11 | 60 | |
| Two-Stream CNNs [18]Pre-trained=ImageNet2026.01 | 59.4 | |
| Swin-S finetune (our baseline)[26]Pre-trained=Kinetics-4002026.01 | 57.47 | |
| XKDPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 57.4 | |
| CrissCrossPretrain=AS, Modality=VA, Protocol=Linear2022.11 | 56.2 | |
| XDCPretrain=AS, Modality=VA, Protocol=Linear2022.11 | 56 | |
| XKD-MASPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 55.1 | |
| AVSlowFastPretrain=K400, Modality=VA, Protocol=Fine-tuned2022.11 | 54.6 | |
| VideoMAEPretrain=K400, Modality=V, Protocol=Linear2022.11 | 54.3 | |
| XKD-MATSPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 53.1 | |
| XDCPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 52.6 | |
| CrissCrossPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 50 | |
| SeLaViPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 47.1 | |
| COCLRPretrain=K400, Modality=VF, Protocol=Linear2022.11 | 46.1 | |
| AVSlowFastPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 44.1 | |
| AVIDPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 41.4 | |
| GDTPretrain=K400, Modality=VA, Protocol=Linear2022.11 | 38.5 |