Human-Object Interaction Detection on V-COCO (mAP, SHOE-mAP)
50.47mAPGPT-4.1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-4.1Confidence score proxy=VLMs token probabilities, Object detector=DETR2026.04 | 50.47 | 57.82 | |
| InternVL3-38BConfidence score proxy=VLMs token probabilities, Object detector=DETR2026.04 | 46.34 | 50.79 | |
| Qwen2.5-VL-32BConfidence score proxy=VLMs token probabilities, Object detector=DETR2026.04 | 41.12 | 64.77 |