Human-Object Interaction Detection on HICO-DET Known Object (test)
51.75mAP (Full)Spatially Conditioned Graphs
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Spatially Conditioned GraphsBackbone=ResNet-50-FPN, Detector Setting=Oracle Detector2020.12 | 51.75 | 41.4 | 54.84 | |
| GRASP-HOIBackbone=ResNet-50+MLLM, Closed-set setting=true2025.12 | 51.57 | 51.53 | 51.61 | |
| HORPBackbone=Swin-L+CLIP, Closed-set setting=true2025.12 | 51.24 | 50.78 | 51.38 | |
| PAFRBackbone=Swin-L, Closed-set setting=true2025.12 | 49.5 | 50.59 | 49.18 | |
| SICHOIBackbone=R101+VIT-L, Closed-set setting=true2025.12 | 48.16 | 48.37 | 48.09 | |
| PVICBackbone=Swin-L, Closed-set setting=true2025.12 | 47.81 | 48.38 | 47.64 | |
| BC-HOIBackbone=ResNet-50, Closed-set setting=true2025.12 | 45.35 | 47.94 | 44.57 | |
| DiffHOIBackbone=Swin-L, Closed-set setting=true2025.12 | 43.62 | 41.41 | 44.28 | |
| UniHOI-l (w/ BLIP2)Extras=L2023.11 | 43.26 | 43.12 | 43.25 | |
| UniHOI-1Backbone=R101+VIT-L, Closed-set setting=true2025.12 | 43.26 | 43.12 | 43.25 | |
| UniHOI-m (w/ BLIP2)Extras=L2023.11 | 42.96 | 42.86 | 42.98 | |
| BCOMBackbone=ResNet-50, Closed-set setting=true2025.12 | 42.24 | 42.86 | 42.05 | |
| UniHOI-s (w/ BLIP2)Extras=L2023.11 | 42.2 | 42.6 | 42.08 | |
| VIPLOBackbone=R50+CLIP, Closed-set setting=true2025.12 | 40.61 | 38.82 | 41.15 | |
| ScriptHOIEvaluation Setting=Known Object2026.05 | 40 | 41.4 | 39.6 | |
| AGERBackbone=ResNet-50, Closed-set setting=true2025.12 | 39.84 | 35.58 | 40.23 | |
| DP-HOIBackbone=ResNet-50, Closed-set setting=true2025.12 | 39.37 | 36.59 | 40.2 | |
| FGAHOIArchitecture=One-Stage / Transformer-Based, Backbone=Swin-Large*, Fine-tuned=false, Feature=A2023.01 | 38.93 | 31.93 | 41.02 | |
| FGAHOIBackbone=Swin-L, Closed-set setting=true2025.12 | 38.93 | 31.93 | 41.02 | |
| CQLBackbone=ResNet-101, Closed-set setting=true2025.12 | 38.82 | 35.51 | 39.81 | |
| RmLRBackbone=ResNet-101, Closed-set setting=true2025.12 | 38.69 | 31.27 | 40.91 | |
| CLIP4HOIEvaluation Setting=Known Object2026.05 | 38 | 35.1 | 38.9 | |
| HOICLIPBackbone=ResNet-50, Training-free enhancement=false2023.03 | 37.96 | 34.49 | 39.02 | |
| GEN-VLKTBackbone=ResNet-502023.03 | 37.8 | 34.76 | 38.71 | |
| HOICLIPBackbone=ResNet-50, Training-free enhancement=true2023.03 | 37.61 | 34.47 | 38.54 | |
| HOICLIPExtras=L2023.11 | 37.61 | 34.47 | 38.54 | |
| HOICLIPBackbone=ResNet-50, Closed-set setting=true2025.12 | 37.61 | 34.47 | 38.54 | |
| QAHOIArchitecture=One-Stage / Transformer-Based, Backbone=Swin-Large*, Fine-tuned=false, Feature=A2023.01 | 37.59 | 31.66 | 39.36 | |
| QAHOI (R)Architecture=One-Stage / Transformer-Based, Backbone=Swin-Large*, Fine-tuned=false, Feature=A2023.01 | 37.23 | 31.01 | 39.09 | |
| GEN-VLKTEvaluation Setting=Known Object2026.05 | 37.2 | 33.8 | 38.2 | |
| CLIP4HOIBackbone=ResNet-50, Closed-set setting=true2025.12 | 37.19 | 35.27 | 37.77 | |
| GEN-VLKTBackbone=R50+CLIP, Closed-set setting=true2025.12 | 36.78 | 32.75 | 37.99 | |
| Liu et al.Backbone=ResNet-502023.03 | 36.28 | 33.16 | 37.21 | |
| UPTBackbone=ResNet-101, Closed-set setting=true2025.12 | 36.08 | 31.41 | 37.47 | |
| MURENBackbone=R50, Feature=A2023.04 | 35.52 | 30.88 | 36.91 | |
| STIPBackbone=R50, Feature=A+S+L, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 35.29 | 31.43 | 36.45 | |
| STIPBackbone=R50, Feature=A+S+L2023.04 | 35.29 | 31.43 | 36.45 | |
| UPTBackbone=R50, Feature=A+S2023.04 | 35.05 | 29.27 | 36.77 | |
| CDNBackbone=R101, Feature=A2023.04 | 34.79 | 29.48 | 36.38 | |
| CDNBackbone=ResNet-502023.03 | 34.53 | 29.73 | 35.96 | |
| CDNBackbone=ResNet-50, Closed-set setting=true2025.12 | 34.53 | 29.73 | 35.96 | |
| DisTRBackbone=R50, Feature=A2023.04 | 34.5 | 30.13 | 35.81 | |
| STIPBackbone=R50, Feature=A, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 34.41 | 30.12 | 35.69 | |
| SCGArchitecture=Two-Stage / Graph-Based, Backbone=ResNet-50-FPN, Fine-tuned=true, Feature=A+S2023.01 | 34.37 | 27.18 | 36.5 | |
| SCGBackbone=R50-FPN, Feature=A+S+M2023.04 | 34.37 | 27.18 | 36.52 | |
| SCGBackbone=ResNet-502023.03 | 34.37 | 27.18 | 36.52 | |
| Spatially Conditioned GraphsBackbone=ResNet-50-FPN, Detector Setting=Fine-tuned on HICO-DET2020.12 | 34.37 | 27.18 | 36.52 | |
| MSTREvaluation Setting=Known Object2026.05 | 34.2 | 29.8 | 35.5 | |
| MSTRBackbone=R50, Feature=A+M2023.04 | 34.02 | 28.83 | 35.57 | |
| CDNEvaluation Setting=Known Object2026.05 | 33.8 | 29.1 | 35.2 | |
| PhraseHOIExtras=L2023.11 | 33.74 | 27.35 | 35.64 | |
| STIPBackbone=R50, Feature=A+S+L, Training Setting=Object detector fine-tuned on HICO-DET2022.06 | 33.54 | 30.93 | 34.31 | |
| GGNetBackbone=HG104, Feature=A, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 33.5 | 26.67 | 34.89 | |
| STIPBackbone=R50, Feature=A, Training Setting=Object detector fine-tuned on HICO-DET2022.06 | 32.84 | 29.05 | 33.85 | |
| FGAHOIArchitecture=One-Stage / Transformer-Based, Backbone=Swin-Tiny, Fine-tuned=false, Feature=A2023.01 | 32.48 | 24.16 | 34.97 | |
| QPICBackbone=R101, Feature=A2023.04 | 32.38 | 26.06 | 34.27 | |
| QPICBackbone=ResNet-101, Closed-set setting=true2025.12 | 32.38 | 26.06 | 34.27 | |
| STIPBackbone=R50, Feature=A+S+L, Training Setting=Object detector pre-trained on MS-COCO2022.06 | 32.28 | 31.07 | 32.64 | |
| AS-NetBackbone=R50, Feature=A, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 31.74 | 27.07 | 33.14 | |
| AS-NetArchitecture=One-Stage / Transformer-Based, Backbone=ResNet-50, Fine-tuned=true, Feature=A, Row=22023.01 | 31.74 | 27.07 | 33.14 | |
| AS-NetBackbone=R50, Feature=A2023.04 | 31.74 | 27.07 | 33.14 | |
| AS-NetBackbone=ResNet-502023.03 | 31.74 | 27.07 | 33.14 | |
| QPICEvaluation Setting=Known Object2026.05 | 31.7 | 24.1 | 34 | |
| QPICBackbone=R50, Feature=A, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 31.68 | 24.14 | 33.93 | |
| QPICArchitecture=One-Stage / Transformer-Based, Backbone=ResNet-50, Fine-tuned=true, Feature=A2023.01 | 31.68 | 24.14 | 33.93 | |
| QPICBackbone=ResNet-502023.03 | 31.68 | 24.14 | 33.93 | |
| FCLBackbone=ResNet-502023.03 | 31.31 | 25.62 | 33.02 | |
| STIPBackbone=R50, Feature=A, Training Setting=Object detector pre-trained on MS-COCO2022.06 | 31.23 | 27.93 | 32.22 | |
| ATLBackbone=ResNet-502023.03 | 31.18 | 24.15 | 33.29 | |
| QAHOIArchitecture=One-Stage / Transformer-Based, Backbone=Swin-Tiny, Fine-tuned=false, Feature=A2023.01 | 30.99 | 24.83 | 32.84 | |
| QAHOI (R)Architecture=One-Stage / Transformer-Based, Backbone=Swin-Tiny, Fine-tuned=false, Feature=A2023.01 | 30.06 | 22.95 | 32.18 | |
| HoiTransBackbone=R101, Feature=A2023.04 | 29.13 | 20.98 | 31.57 | |
| HOITransArchitecture=One-Stage / Transformer-Based, Backbone=ResNet-101, Fine-tuned=true, Feature=A2023.01 | 29.1 | 20.98 | 31.57 | |
| IDNBackbone=R50, Feature=A+S, Training Setting=Object detector fine-tuned on HICO-DET2022.06 | 28.24 | 24.47 | 29.37 | |
| IDNBackbone=R50, Feature=A+S2023.04 | 28.24 | 24.47 | 29.37 | |
| IDNBackbone=ResNet50, Detector Setting=Fine-tuned on HICO-DET2020.12 | 28.24 | 24.47 | 29.37 | |
| DRGBackbone=R50-FPN, Feature=A+S+L, Training Setting=Object detector fine-tuned on HICO-DET2022.06 | 27.98 | 23.11 | 29.43 | |
| DRGArchitecture=Two-Stage / Graph-Based, Backbone=ResNet-50-FPN, Fine-tuned=true, Feature=A+S+L2023.01 | 27.98 | 23.14 | 29.43 | |
| DRGBackbone=R50-FPN, Feature=A+S+L+M2023.04 | 27.98 | 23.11 | 29.43 | |
| DRGExtras=P2023.11 | 27.98 | 23.11 | 29.43 | |
| DRGBackbone=ResNet-50-FPN, Detector Setting=Fine-tuned on HICO-DET2020.12 | 27.98 | 23.11 | 29.43 | |
| AS-NetBackbone=R50, Feature=A, Training Setting=Object detector pre-trained on MS-COCO2022.06 | 27.41 | 25.44 | 28 | |
| AS-NetArchitecture=One-Stage / Transformer-Based, Backbone=ResNet-50, Fine-tuned=true, Feature=A, Row=12023.01 | 27.41 | 25.44 | 28 | |
| GGNetArchitecture=One-Stage / Interaction points, Backbone=Hourglass-104, Fine-tuned=true, Feature=A2023.01 | 27.36 | 20.23 | 29.48 | |
| GGNetBackbone=HG104, Feature=A2023.04 | 27.36 | 20.23 | 29.48 | |
| PD-NetArchitecture=Two-Stage / Multi-stream, Backbone=ResNet-152, Fine-tuned=true, Feature=A+S+P+L2023.01 | 26.86 | 21.7 | 28.44 | |
| IDNBackbone=R50, Feature=A+S, Training Setting=Object detector pre-trained on MS-COCO2022.06 | 26.43 | 25.01 | 26.85 | |
| IDNBackbone=ResNet-502023.03 | 26.43 | 25.01 | 26.85 | |
| IDNBackbone=ResNet50, Detector Setting=Pre-trained on MS COCO2020.12 | 26.43 | 25.01 | 26.85 | |
| PSTBackbone=R50, Feature=A2023.04 | 26.42 | 17.61 | 29.05 | |
| HOITransBackbone=R50, Feature=A, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 26.15 | 19.24 | 28.22 | |
| Zou et al.Backbone=ResNet-502023.03 | 26.15 | 19.24 | 28.22 | |
| VCLArchitecture=Two-Stage / Multi-stream, Backbone=ResNet-50, Fine-tuned=true, Feature=A+S2023.01 | 25.98 | 19.12 | 28.03 | |
| VCLBackbone=R50, Feature=A2023.04 | 25.98 | 19.12 | 28.03 | |
| VCLBackbone=ResNet50, Detector Setting=Fine-tuned on HICO-DET2020.12 | 25.98 | 19.12 | 28.03 | |
| PD-NetExtras=L2023.11 | 25.59 | 19.93 | 27.28 | |
| Spatially Conditioned GraphsBackbone=ResNet-50-FPN, Detector Setting=Pre-trained on MS COCO2020.12 | 25.53 | 21.79 | 26.64 | |
| PPDMBackbone=HG104, Feature=A, Training Setting=Jointly fine-tune object detector & HOI detector2022.06 | 24.58 | 16.65 | 26.84 | |
| PPDMArchitecture=One-Stage / Interaction points, Backbone=Hourglass-104, Fine-tuned=true, Feature=A2023.01 | 24.58 | 16.65 | 26.84 | |
| PPDMBackbone=ResNet-502023.03 | 24.58 | 16.65 | 26.84 |