Scene Graph Generation on PARSE-10K (test)
73.2Recall (BBox Match)Ours
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| OursEvaluation protocol=Fine-tuned, Base Model=Qwen3-VL2026.03 | 73.2 | 74.8 | 80.3 | 82 | 76.6 | 78.2 | 14.1 | |
| Gemini-2.5-ProEvaluation protocol=Zero-shot2026.03 | 40.5 | 43.4 | 48.6 | 52 | 44.2 | 47.3 | 12.9 | |
| Qwen3-VLEvaluation protocol=Zero-shot2026.03 | 26 | 29.6 | 46 | 52.4 | 33.2 | 37.9 | 8.7 | |
| GPT-5Evaluation protocol=Zero-shot2026.03 | 13.7 | 40.9 | 13.9 | 41.3 | 13.8 | 41.1 | 15.3 | |
| Robobrain2.0Evaluation protocol=Zero-shot2026.03 | 9.2 | 11.3 | 26.7 | 32.8 | 13.7 | 16.9 | 5.6 | |
| Claude-Opus-4Evaluation protocol=Zero-shot2026.03 | 8 | 33.7 | 12.7 | 53.7 | 9.8 | 41.4 | 9.7 |