Cross-view Temporal Reasoning on EgoExo-Con V-Ego
74.4AccuracyInternVL3.5-8B + View-GRPO
Evaluation results
| Method | Links | |
|---|---|---|
| InternVL3.5-8B + View-GRPOBackbone=InternVL3.5-8B, Training Protocol=View-GRPO2025.10 | 74.4 | |
| InternVL3.5-8BBackbone=InternVL3.5-8B, Training Protocol=Base2025.10 | 64.7 | |
| Qwen2.5-VL-7B + View-GRPOBackbone=Qwen2.5-VL-7B, Training Protocol=View-GRPO2025.10 | 58.2 | |
| Qwen2.5-VL-7B + SFTBackbone=Qwen2.5-VL-7B, Training Protocol=SFT2025.10 | 58 | |
| Qwen2.5-VL-7B + GRPOBackbone=Qwen2.5-VL-7B, Training Protocol=GRPO2025.10 | 57.6 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Training Protocol=Base2025.10 | 56.3 | |
| Qwen2.5-VL-3B + View-GRPOBackbone=Qwen2.5-VL-3B, Training Protocol=View-GRPO2025.10 | 54.4 | |
| Qwen2.5-VL-3B + GRPOBackbone=Qwen2.5-VL-3B, Training Protocol=GRPO2025.10 | 52.9 | |
| Qwen2.5-VL-3BBackbone=Qwen2.5-VL-3B, Training Protocol=Base2025.10 | 52.5 | |
| Qwen2.5-VL-3B + SFTBackbone=Qwen2.5-VL-3B, Training Protocol=SFT2025.10 | 51.2 |