Pedestrian Detection on CrowdHuman (val)
36.5MR^-2Hulk
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| HulkBackbone=ViT-L, Fine-tuning=true, Input Resolution=11202023.12 | 36.5 | 93 | 87 | — | — | — | — | |
| Iter-DDETRBackbone=Swin-L2023.12 | 37.7 | 94.1 | 87.1 | — | — | — | — | |
| CascadeR-CNN + Oursdetector=Cascade R-CNN2020.03 | 38.7 | — | 83.9 | — | — | — | 90.6 | |
| HulkBackbone=ViT-L, Fine-tuning=false, Input Resolution=11202023.12 | 40.1 | 92.2 | 85.8 | — | — | — | — | |
| CascadeR-CNNimplementation=our implementation2020.03 | 40.2 | — | 80.4 | — | — | — | 86.2 | |
| Beta R-CNNKLth=72022.10 | 40.3 | — | — | — | — | — | 88.2 | |
| Beta R-CNNKLth=62022.10 | 40.6 | — | — | — | — | — | 89.6 | |
| HulkBackbone=ViT-B, Fine-tuning=true, Input Resolution=11202023.12 | 40.7 | 92.4 | 86 | — | — | — | — | |
| CrowdDetBackbone=ResNet-502023.12 | 41.4 | 90.7 | 82.3 | — | — | — | — | |
| Oursdetector=FPN2020.03 | 41.4 | — | 82.3 | — | — | — | 90.7 | |
| CrowdDetBackbone=Res-50 + FPN, Image size=Standard2021.04 | 41.4 | — | — | 83.68 | — | — | 90.7 | |
| CrowdDet2022.10 | 41.4 | — | — | — | — | — | 90.7 | |
| Iter-DDETRBackbone=ResNet-502023.12 | 41.5 | 92.1 | 84 | — | — | — | — | |
| Iter-Deformable-DETR2023.03 | 41.5 | 92.1 | 84 | — | — | — | — | |
| UniHCPBackbone=ViT-B, Fine-tuning=true2023.12 | 41.6 | 92.5 | 85.8 | — | — | — | — | |
| Iter-Sparse-RCNN2023.03 | 41.6 | 92.5 | 83.3 | — | — | — | — | |
| UniHCPEvaluation protocol=finetune2023.03 | 41.6 | 92.5 | 85.8 | — | — | — | — | |
| V2F-NetBackbone=Res-50 + FPN, Image size=Standard2021.04 | 42.28 | — | — | 84.2 | — | — | 91.03 | |
| V2F-NetBackbone=ResNet-502023.12 | 42.3 | 91 | — | — | — | — | — | |
| Iter-Sparse-RCNNBackbone=ResNet-502023.12 | 42.6 | 92.5 | 83.3 | — | — | — | — | |
| FPN baselinedetector=FPN2020.03 | 42.9 | — | 79.8 | — | — | — | 85.8 | |
| FPN + Soft-NMSdetector=FPN2020.03 | 42.9 | — | 79.8 | — | — | — | 88.2 | |
| PEDRBackbone=ResNet-502023.12 | 43.3 | 91.6 | 83.3 | — | — | — | — | |
| PBM2022.10 | 43.3 | — | — | — | — | — | 89.3 | |
| R2NMSBackbone=Res-50 + FPN, Image size=Standard2021.04 | 43.35 | — | — | — | — | — | 89.29 | |
| DDETRBackbone=ResNet-502023.12 | 43.7 | 91.5 | 83.1 | — | — | — | — | |
| PEDR2023.03 | 43.7 | 91.6 | 83.3 | — | — | — | — | |
| Deformable-DETR2023.03 | 43.7 | 91.5 | 83.1 | — | — | — | — | |
| HulkBackbone=ViT-B, Fine-tuning=false, Input Resolution=11202023.12 | 43.8 | 90.7 | 84 | — | — | — | — | |
| NOH-NMSBackbone=ResNet-50, Extra Annotations=false2020.07 | 43.9 | — | — | 92.9 | — | — | 89 | |
| NOH-NMSBackbone=Res-50 + FPN, Image size=Standard2021.04 | 43.9 | — | — | — | — | — | 89 | |
| Our BaselineBackbone=ResNet-50, Extra Annotations=false2020.07 | 44.7 | — | — | 87.8 | — | — | 85.1 | |
| Sparse-RCNNBackbone=ResNet-502023.12 | 44.8 | 91.3 | 81.3 | — | — | — | — | |
| Sparse-RCNN2023.03 | 44.8 | 91.3 | 81.3 | — | — | — | — | |
| Repulsion LossBackbone=ResNet-50, Extra Annotations=false2020.07 | 45.7 | — | — | — | — | — | — | |
| JointDetBackbone=ResNet-50, Extra Annotations=true2020.07 | 46.5 | — | — | — | — | — | — | |
| UniHCPBackbone=ViT-B, Fine-tuning=false2023.12 | 46.6 | 90 | 82.2 | — | — | — | — | |
| UniHCPEvaluation protocol=direct eval2023.03 | 46.6 | 90 | 82.2 | — | — | — | — | |
| BaselineBackbone=Res-50 + FPN, Image size=Standard2021.04 | 46.95 | — | — | 76.9 | — | — | 85.18 | |
| CaSeBackbone=Res-50 + FPN, Image size=Standard2021.04 | 47.9 | — | — | — | — | — | — | |
| RelationNetdetector=FPN, implementation=our implementation2020.03 | 48.2 | — | 74.6 | — | — | — | 81.6 | |
| GossipNetdetector=FPN, implementation=our implementation2020.03 | 49.1 | — | 81.6 | — | — | — | 80.4 | |
| AdaptiveNMSimplementation=impl. by [23]2020.03 | 49.7 | — | — | — | — | — | 84.7 | |
| Adaptive-NMSBackbone=ResNet-50, Extra Annotations=false2020.07 | 49.7 | — | — | 91.3 | — | — | 84.7 | |
| Adaptive NMS2022.10 | 49.7 | — | — | — | — | — | 84.7 | |
| FPNNMS=adaptive2019.04 | 49.73 | 84.71 | — | 91.27 | — | — | — | |
| Adaptive-NMSBackbone=Res-50 + FPN, Image size=Standard2021.04 | 49.73 | — | — | — | — | — | 84.71 | |
| CrowdHuman2022.10 | 50.4 | — | — | — | — | — | 85 | |
| FPNNMS=greedy, Reference=[36]2019.04 | 50.42 | 84.95 | — | 90.24 | — | — | — | |
| FPNNMS=soft2019.04 | 51.97 | 83.99 | — | 91.73 | — | — | — | |
| FPNNMS=greedy2019.04 | 52.35 | 83.07 | — | 90.57 | — | — | — | |
| FPN baselineimplementation=impl. by [23]2020.03 | 52.4 | — | — | — | — | — | 83.1 | |
| Baseline in [18]Backbone=ResNet-50, Extra Annotations=false2020.07 | 52.4 | — | — | 90.6 | — | — | 83 | |
| CSP2019.04 | 62.1 | — | — | — | — | — | — | |
| RFB NetNMS=adaptive2019.04 | 63.03 | 79.67 | — | 94.77 | — | — | — | |
| RetinaNet2019.04 | 63.3 | — | — | — | — | — | — | |
| RetinaNetNMS=greedy, Reference=[36]2019.04 | 63.33 | 80.83 | — | 93.8 | — | — | — | |
| RFB NetNMS=greedy2019.04 | 65.22 | 78.33 | — | 94.13 | — | — | — | |
| RFB NetNMS=soft2019.04 | 66.34 | 78.1 | — | 95.37 | — | — | — | |
| DETRBackbone=ResNet-502023.12 | 73.2 | 75.9 | 74.4 | — | — | — | — | |
| DETR2023.03 | 73.2 | 75.9 | 74.4 | — | — | — | — | |
| Adaptive NMS2020.03 | — | 84.71 | — | 91.27 | -0.31 | 1.03 | — | |
| BaselineRegion Feature Extraction=RoI Align2020.03 | — | 85.02 | — | 90.62 | — | — | — | |
| Baseline Shao et al.Region Feature Extraction=RoI Pooling [4]2020.03 | — | 84.95 | — | 90.24 | -0.07 | -0.38 | — | |
| PATHBackbone=ViT-B2023.12 | — | 90.9 | — | — | — | — | — | |
| PATHBackbone=ViT-L2023.12 | — | 90.8 | — | — | — | — | — | |
| PS-RCNNRegion Feature Extraction=RoI Align; RoI Align2020.03 | — | 85.53 | — | 92.67 | 0.51 | 2.05 | — | |
| PS-RCNNRegion Feature Extraction=RoI Align; HRRA2020.03 | — | 86.05 | — | 93.77 | 1.03 | 3.15 | — | |
| PS-RCNN w/ IMRegion Feature Extraction=RoI Align; HRRA2020.03 | — | 87.94 | — | 95.11 | 2.92 | 4.49 | — | |
| Repulsion LossRegion Feature Extraction=RoI Align2020.03 | — | 85.71 | — | 90.74 | 0.69 | 0.12 | — | |
| Soft-NMSRegion Feature Extraction=RoI Align2020.03 | — | 85.66 | — | 92.88 | 0.64 | 2.26 | — |