Video reasoning on Video-R1
44.3VSIAT-RL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AT-RLBackbone=Qwen2.5-VL-7B, Frames=642026.02 | 44.3 | 56 | 70.2 | |
| AT-RLBackbone=Qwen2.5-VL-7B, Frames=322026.02 | 39.6 | 53.4 | 67.2 | |
| T-GRPOBackbone=Qwen2.5-VL-7B, Frames=642026.02 | 39.2 | 50.8 | 65 | |
| SFTBackbone=Qwen2.5-VL-7B, Frames=642026.02 | 36.8 | 51 | 63.5 | |
| T-GRPOBackbone=Qwen2.5-VL-7B, Frames=322026.02 | 36.5 | 50.2 | 64 | |
| AT-RLBackbone=Qwen2.5-VL-7B, Frames=162026.02 | 36.4 | 51.5 | 64.8 | |
| SFTBackbone=Qwen2.5-VL-7B, Frames=322026.02 | 34.2 | 50.5 | 62.3 | |
| T-GRPOBackbone=Qwen2.5-VL-7B, Frames=162026.02 | 33.8 | 48.9 | 62.1 | |
| Zero-shotBackbone=Qwen2.5-VL-7B, Frames=642026.02 | 32.5 | 48.5 | 61.2 | |
| SFTBackbone=Qwen2.5-VL-7B, Frames=162026.02 | 31.5 | 49.2 | 60.8 | |
| Zero-shotBackbone=Qwen2.5-VL-7B, Frames=322026.02 | 30.1 | 48.1 | 60 | |
| Zero-shotBackbone=Qwen2.5-VL-7B, Frames=162026.02 | 27.7 | 47.8 | 59.2 |