Human-Object Interaction Detection on HICO-DET (single HOI subset 575 classes standard test)
49.5mAPGPT-4.1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4.1Method Type=DETR + VLMs, Detector=DETR2026.04 | 49.5 | 61.67 | |
| InternVL3-38BMethod Type=DETR + VLMs, Detector=DETR2026.04 | 42 | 58.03 | |
| HOLAMethod Type=Default, Backbone=ViT-L2026.04 | 39.05 | 39.92 | |
| ADA-CMMethod Type=Default, Backbone=ViT-L2026.04 | 38.4 | 39.72 | |
| CMMPMethod Type=Default, Backbone=ViT-L2026.04 | 37.77 | 39.06 | |
| LAINMethod Type=Default, Backbone=ViT-B2026.04 | 35.3 | 36.58 | |
| Qwen2.5-VL-32BMethod Type=DETR + VLMs, Detector=DETR2026.04 | 34.83 | 66.03 | |
| LAINMethod Type=Zero-shot RF-UC, Backbone=ViT-B2026.04 | 34.6 | 35.37 | |
| HOLAMethod Type=Zero-shot RF-UC, Backbone=ViT-B2026.04 | 34.19 | 35.26 | |
| EZ-HOIMethod Type=Zero-shot RF-UC, Backbone=ViT-B2026.04 | 33.13 | 33.83 | |
| CMMPMethod Type=Zero-shot RF-UC, Backbone=ViT-B2026.04 | 32.18 | 32.95 | |
| THIDMethod Type=Open Vocabulary2026.04 | 22.01 | 22.04 | |
| CMD-SEMethod Type=Open Vocabulary2026.04 | 18.2 | 19.03 |