Triple-Shot Composition on TSC-Bench
54.4IoUShotCrop3
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ShotCrop3Category=Ours2026.06 | 54.4 | 8.7 | 55.4 | 60 | 45.5 | 82.6 | 62.3 | |
| Qwen3-VL-4BCategory=4b-Scale MLLMs, sft=true2026.06 | 45.7 | 10.6 | 54.3 | 59.3 | 44.9 | 81.2 | 57.8 | |
| InternVL3.5-4BCategory=4b-Scale MLLMs, sft=true2026.06 | 42.4 | 12 | 54.7 | 59.3 | 44.8 | 81.3 | 58.2 | |
| Qwen3-VL-32BCategory=Large-Scale MLLMs2026.06 | 31.2 | 15 | 51.1 | 56 | 41.5 | 80.9 | 58 | |
| Qwen3-VL-4Category=4b-Scale MLLMs2026.06 | 29.2 | 15.8 | 50.5 | 55.1 | 41.2 | 79.3 | 52.1 | |
| DeepPerceptionCategory=Specialist Models2026.06 | 26.8 | 16.8 | 49.9 | 54.7 | 40.5 | 78.8 | 51.1 | |
| InstructCropCategory=Specialist Models2026.06 | 25.2 | 18 | 54.4 | 58.9 | 45.1 | 82.4 | 37.1 | |
| InternVL3.5-38BCategory=Large-Scale MLLMs2026.06 | 23.8 | 17 | 53.2 | 56.2 | 41.5 | 80.8 | 61.7 | |
| Gemini2.5 proCategory=Closed-source Models2026.06 | 18.7 | 24.6 | 54.7 | 57.5 | 43.7 | 81.5 | 62.1 | |
| gpt-5Category=Closed-source Models2026.06 | 16.8 | 23.2 | 55 | 58.6 | 43.8 | 81 | 60.9 | |
| InternVL3.5-4BCategory=4b-Scale MLLMs2026.06 | 14.1 | 20.4 | 49 | 50.3 | 36.7 | 72.3 | 46.5 |