Visual Evidence Recoverability Evaluation on VidCapQA 1.0 (test)
57.6Overall AccuracySeed 2.0 Pro
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Seed 2.0 ProConfiguration=+sec2026.05 | 57.6 | 35.7 | 61.1 | 58.3 | 66.7 | 26.9 | 47.8 | 74.3 | 48.6 | 63.9 | 73.2 | 69 | 48.6 | 76.1 | 53.5 | |
| Seed 2.0 ProDescription=Seed2026.05 | 50.9 | 25.7 | 54.9 | 51.4 | 66.2 | 14.5 | 29 | 65.1 | 42.2 | 58.7 | 73.1 | 78.6 | 41 | 69 | 44.4 | |
| VDCTalkerDescription=Ours2026.05 | 49.5 | 31 | 62 | 45.8 | 51.4 | 23.2 | 26.8 | 69 | 45.1 | 62.5 | 62 | 73.2 | 38.1 | 59.1 | 41.7 | |
| Gemini-3.1 Pro-Preview2026.05 | 47.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3.5-35B-A3BDescription=Base model2026.05 | 44.4 | 15.5 | 50 | 33.8 | 42.2 | 16.9 | 21.1 | 64.2 | 42.2 | 62 | 59.1 | 71.8 | 36.9 | 60.6 | 45.8 |