Pedestrian Detection on CrowdHuman
97.2mAPInternImage-H
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| InternImage-H#Param.=2.18B, Composite techniques [30]=true2024.07 | 97.2 | 31.1 | 89.7 | |
| MMPedestron#Param.=62M2024.07 | 97.1 | 30.8 | 88 | |
| InternImage-H#Param.=1.09B2024.07 | 95.4 | 37.9 | 86.6 | |
| Iter-DDETRBackbone=Swin-L2023.12 | 94.1 | 37.7 | 87.1 | |
| Iter-D-DETR#Param.=207M2024.07 | 94.1 | 37.7 | 87.1 | |
| Hulk-FTBackbone=ViT-L, Evaluation Protocol=Fine-tuned, Input Resolution=Smaller (Max Height/Width 1120)2023.12 | 93 | 36.5 | 87 | |
| Iter-Sparse-RCNNBackbone=ResNet-502023.12 | 92.5 | 42.6 | 83.3 | |
| UniHCP-FTBackbone=ViT-B, Evaluation Protocol=Fine-tuned2023.12 | 92.5 | 41.6 | 85.8 | |
| SoTA †Note=multitask or stronger models2023.03 | 92.5 | — | — | |
| UniHCP#Param.=109M2024.07 | 92.5 | 41.6 | 85.8 | |
| Hulk-FTBackbone=ViT-B, Evaluation Protocol=Fine-tuned, Input Resolution=Smaller (Max Height/Width 1120)2023.12 | 92.4 | 40.7 | 86 | |
| HulkBackbone=ViT-L, Evaluation Protocol=Direct Eval, Input Resolution=Smaller (Max Height/Width 1120)2023.12 | 92.2 | 40.1 | 85.8 | |
| Iter-DDETRBackbone=ResNet-502023.12 | 92.1 | 41.5 | 84 | |
| SoTA2023.03 | 92.1 | — | — | |
| PEDRBackbone=ResNet-502023.12 | 91.6 | 43.7 | 83.3 | |
| PEDR#Param.=41M2024.07 | 91.6 | 43.7 | 83.3 | |
| DDETRBackbone=ResNet-502023.12 | 91.5 | 43.7 | 83.1 | |
| D-DETR#Param.=40M2024.07 | 91.5 | 43.7 | 83.1 | |
| Sparse-RCNNBackbone=ResNet-502023.12 | 91.3 | 44.8 | 81.3 | |
| S-RCNN#Param.=106M2024.07 | 91.3 | 44.8 | 81.3 | |
| V2F-NetBackbone=ResNet-502023.12 | 91 | 42.3 | — | |
| PATHBackbone=ViT-B, Evaluation Protocol=Pretraining2023.12 | 90.9 | — | — | |
| PATHBackbone=ViT-B, Evaluation Protocol=Partial FT2023.03 | 90.9 | — | — | |
| PATHBackbone=ViT-L, Evaluation Protocol=Pretraining2023.12 | 90.8 | — | — | |
| PATHBackbone=ViT-L, Evaluation Protocol=Partial FT2023.03 | 90.8 | — | — | |
| PATH#Param.=320M2024.07 | 90.8 | — | — | |
| CrowdDetBackbone=ResNet-502023.12 | 90.7 | 41.4 | 82.3 | |
| HulkBackbone=ViT-B, Evaluation Protocol=Direct Eval, Input Resolution=Smaller (Max Height/Width 1120)2023.12 | 90.7 | 43.8 | 84 | |
| CrowdDet#Param.=42M2024.07 | 90.7 | 41.4 | 82.3 | |
| PATHBackbone=ViT-B, Evaluation Protocol=FT2023.03 | 90.6 | — | — | |
| UniHCPBackbone=ViT-B, Evaluation Protocol=Direct Eval2023.12 | 90 | 46.6 | 82.2 | |
| PATHBackbone=ViT-B, Evaluation Protocol=Head FT2023.03 | 90 | — | — | |
| MAEBackbone=ViT-B2023.03 | 89.6 | — | — | |
| PATHBackbone=ViT-L, Evaluation Protocol=w/o FT2023.03 | 89.4 | — | — | |
| PATHBackbone=ViT-B, Evaluation Protocol=w/o FT2023.03 | 89.1 | — | — | |
| CLIPBackbone=ViT-B2023.03 | 82.1 | — | — | |
| DETRBackbone=ResNet-502023.12 | 75.9 | 73.2 | 74.4 | |
| DETR#Param.=41M2024.07 | 75.9 | 73.2 | 74.4 |