Face Detection on WIDER FACE (val)
97.2mAP (Easy)RefineFace
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| RefineFaceBackbone=ResNet-152, TTA=true2020.11 | 97.2 | 96.2 | 92 | — | — | — | — | — | |
| ASFD-D6Backbone=ResNet-152, TTA=true2020.11 | 97.2 | 96.5 | 92.5 | — | — | — | — | — | |
| ASFD-D6Pose?=No2020.12 | 97.2 | 96.5 | 92.5 | — | — | — | — | — | |
| RetinaFaceBackbone=R-152, Pose?=Yes (requires PnP)2020.12 | 97.1 | 96.2 | 92 | — | — | — | — | — | |
| AInnoFaceBackbone=ResNet-152, TTA=true2020.11 | 97 | 96.1 | 91.8 | — | — | — | — | — | |
| HAMBoxBackbone=ResNet-50, TTA=true2020.11 | 97 | 96.4 | 93.3 | — | — | — | — | — | |
| TinaFaceBackbone=ResNet-50, TTA=true2020.11 | 97 | 96.3 | 93.4 | — | — | — | — | — | |
| RetinaFaceBackbone=ResNet-152, TTA=true2020.11 | 96.9 | 96.1 | 91.8 | — | — | — | — | — | |
| DSFD-ResNet152Backbone=ResNet152, # Params=459 M2019.06 | 96.6 | 95.7 | 90.4 | — | — | — | — | — | |
| No AnonymizationFace detector=DSFD [14], Anonymization method=None2019.09 | 96.6 | 95.7 | 90.4 | — | — | — | — | — | |
| DSFDBackbone=R-50, Pose?=No2020.12 | 96.6 | 95.7 | 90.4 | — | — | — | — | — | |
| PyramidBox++Backbone=R-50, Pose?=No2020.12 | 96.5 | 95.9 | 91.2 | — | — | — | — | — | |
| SRNBackbone=R-50, Pose?=No2020.12 | 96.4 | 95.3 | 90.2 | — | — | — | — | — | |
| DSFD-ResNet101Backbone=ResNet101, # Params=399 M2019.06 | 96.3 | 95.4 | 90.1 | — | — | — | — | — | |
| TinaFaceBackbone=ResNet-50, TTA=false2020.11 | 96.3 | 95.7 | 93 | — | — | — | — | — | |
| PyramidBoxBackbone=VGG-16, # Params=57 M, # Madds (G)=1292019.06 | 96.1 | 95 | 88.7 | — | — | — | — | — | |
| SCRFD-34GFBackbone=Bottleneck Res, #Params(M)=9.80, #Flops(G)=34.13, Infer(ms)=11.7, Test Resolution=640x4802021.05 | 96.06 | 94.92 | 85.29 | — | — | — | — | — | |
| DeepPrivacyFace detector=DSFD [14], Anonymization method=DeepPrivacy2019.09 | 95.9 | 95 | 89.8 | — | — | — | — | — | |
| ResNet-34GFBackbone=ResNet50, #Params(M)=24.81, #Flops(G)=34.16, Infer(ms)=11.8, Sample Redistribution=True, Test Resolution=640x4802021.05 | 95.64 | 94.22 | 84.02 | — | — | — | — | — | |
| TinaFaceBackbone=ResNet50, #Params(M)=37.98, #Flops(G)=172.95, Infer(ms)=38.9, Test Resolution=640x4802021.05 | 95.61 | 94.25 | 81.43 | — | — | — | — | — | |
| Pixelation (16x16)Face detector=DSFD [14], Anonymization method=Pixelation, Pixelation scale=16x162019.09 | 95.3 | 94.9 | 90.2 | — | — | — | — | — | |
| 9x9 Gaussian BlurFace detector=DSFD [14], Anonymization method=Gaussian Blur, sigma=3, Kernel size=9x92019.09 | 95.3 | 92.8 | 84.7 | — | — | — | — | — | |
| HAMBoxBackbone=ResNet50, #Params(M)=30.24, #Flops(G)=43.28, Infer(ms)=25.9, Test Resolution=640x4802021.05 | 95.27 | 93.76 | 76.75 | — | — | — | — | — | |
| SCRFD-10GFBackbone=Basic Res, #Params(M)=3.86, #Flops(G)=9.98, Infer(ms)=4.9, Test Resolution=640x4802021.05 | 95.16 | 93.87 | 83.05 | — | — | — | — | — | |
| RetinaFaceBackbone=ResNet50, #Params(M)=29.50, #Flops(G)=37.59, Infer(ms)=21.7, Test Resolution=640x4802021.05 | 94.92 | 91.9 | 64.17 | — | — | — | — | — | |
| ResNet-10GFBackbone=ResNet34x0.5, #Params(M)=6.85, #Flops(G)=10.18, Infer(ms)=6.3, Sample Redistribution=True, Test Resolution=640x4802021.05 | 94.69 | 92.9 | 80.42 | — | — | — | — | — | |
| DSFDBackbone=ResNet152, #Params(M)=120.06, #Flops(G)=259.55, Infer(ms)=55.6, Test Resolution=640x4802021.05 | 94.29 | 91.47 | 71.39 | — | — | — | — | — | |
| S3FDBackbone=VGG-16, # Params=22 M, # Madds (G)=1282019.06 | 94.2 | 93 | 85.9 | — | — | — | — | — | |
| SCRFD-2.5GFBackbone=Basic Res, #Params(M)=0.67, #Flops(G)=2.53, Infer(ms)=4.2, Test Resolution=640x4802021.05 | 93.78 | 92.16 | 77.87 | — | — | — | — | — | |
| ResNet-2.5GFBackbone=ResNet34x0.25, #Params(M)=1.62, #Flops(G)=2.57, Infer(ms)=5.4, Sample Redistribution=True, Test Resolution=640x4802021.05 | 93.21 | 91.11 | 74.47 | — | — | — | — | — | |
| S3FD - ScratchBackbone=VGG-16, # Params=22 M, # Madds (G)=1282019.06 | 93.1 | 92 | 84.6 | — | — | — | — | — | |
| EXTD-FPN-64-PReLU# Params=0.16 M, # Madds (G)=11.22019.06 | 92.1 | 91.1 | 85.6 | — | — | — | — | — | |
| RNNPool-Face-CPeak RAM=6.44 MB, Parameters=1.52M, MAdds=1.80G, RNNPoolLayer hidden dimension=162020.02 | 92 | 89 | 70 | — | — | — | — | — | |
| Pixelation (8x8)Face detector=DSFD [14], Anonymization method=Pixelation, Pixelation scale=8x82019.09 | 91.4 | 92.3 | 88.9 | — | — | — | — | — | |
| EXTD-FPN-48-PReLU# Params=0.10 M, # Madds (G)=6.672019.06 | 91.3 | 90.4 | 84.7 | — | — | — | — | — | |
| LFFDPeak RAM=18.75 MB, Parameters=2.15M, MAdds=9.25G2020.02 | 91 | 88 | 77 | — | — | — | — | — | |
| LFFDPose?=No2020.12 | 91 | 88.1 | 78 | — | — | — | — | — | |
| img2poseBackbone=R-18, Pose?=Yes2020.12 | 90.8 | 89.9 | 84.7 | — | — | — | — | — | |
| RetinaFace-MBackbone=MobileNet, Pose?=Yes (requires PnP)2020.12 | 90.7 | 88.2 | 73.8 | — | — | — | — | — | |
| SCRFD-0.5GFBackbone=Depth-wise Conv, #Params(M)=0.57, #Flops(G)=0.508, Infer(ms)=3.6, Test Resolution=640x4802021.05 | 90.57 | 88.12 | 68.51 | — | — | — | — | — | |
| MobileNet-0.5GFBackbone=MobileNetx0.25, #Params(M)=0.37, #Flops(G)=0.507, Infer(ms)=3.7, Sample Redistribution=True, Test Resolution=640x4802021.05 | 90.38 | 87.05 | 66.68 | — | — | — | — | — | |
| ASFD-D0Pose?=No2020.12 | 90.1 | 87.5 | 74.4 | — | — | — | — | — | |
| EXTDPeak RAM=18.75 MB, Parameters=0.07M, MAdds=8.49G, Variant=EXTD-322020.02 | 90 | 88 | 82 | — | — | — | — | — | |
| EXTD-FPN-32-PReLU# Params=0.063 M, # Madds (G)=4.522019.06 | 89.6 | 88.5 | 82.5 | — | — | — | — | — | |
| S3FD + MobileFaceNetBackbone=MobileFaceNet, # Params=1.2 M, # Madds (G)=12.72019.06 | 88.1 | 85.9 | 74.1 | — | — | — | — | — | |
| FaceboxesPose?=No2020.12 | 87.9 | 85.7 | 77.1 | — | — | — | — | — | |
| RetinaFaceBackbone=MobileNet0.25, #Params(M)=0.44, #Flops(G)=0.802, Infer(ms)=7.9, Test Resolution=640x4802021.05 | 87.78 | 81.16 | 47.32 | — | — | — | — | — | |
| RNNPool-Face-BPeak RAM=1.76 MB, Parameters=1.12M, MAdds=1.18G, RNNPoolLayer hidden dimension=62020.02 | 87 | 84 | 67 | — | — | — | — | — | |
| FaceBoxesPeak RAM=1.76 MB, Parameters=1.01M, MAdds=2.84G2020.02 | 84 | 77 | 39 | — | — | — | — | — | |
| Heavy BlurFace detector=DSFD [14], Anonymization method=Heavy Blur, filter size=30% face width2019.09 | 83.4 | 86.3 | 86.1 | — | — | — | — | — | |
| RNNPool-Face-QuantPeak RAM=225 KB, Parameters=0.07M, MAdds=0.12G, Quantization=byte quantization, RNNPoolLayer hidden dimension=42020.02 | 80 | 78 | 53 | — | — | — | — | — | |
| RNNPool-Face-APeak RAM=1.17 MB, Parameters=0.06M, MAdds=0.10G, RNNPoolLayer hidden dimension=42020.02 | 77 | 75 | 53 | — | — | — | — | — | |
| FaceBoxesBackbone=Not specified, #Params(M)=1.01, #Flops(G)=0.275, Infer(ms)=2.5, Test Resolution=640x4802021.05 | 76.17 | 57.17 | 24.18 | — | — | — | — | — | |
| EagleEyePeak RAM=1.17 MB, Parameters=0.23M, MAdds=0.08G2020.02 | 74 | 70 | 44 | — | — | — | — | — | |
| ASFD-D6Single Path=R101-FAE-FA-FA-Hx4-256, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 66.2 | 63.5 | 52.3 | — | 97.2 | 96.5 | 92.5 | 36.1 | |
| ASFD-D5Single Path=R101-FAE-Hx4-256, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 65.9 | 63.3 | 51.7 | — | 97 | 96.5 | 91.9 | 29.6 | |
| ASFD-D4Single Path=R50-FAE-Hx4-256, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 65.8 | 63.3 | 50.9 | — | 97 | 96.3 | 91.2 | 26.2 | |
| ASFD-D3Single Path=R50-FAE-Hx3-192, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 65.1 | 62.8 | 50.4 | — | 96.6 | 95.9 | 90.5 | 16.6 | |
| ASFD-D2Single Path=R34-FA-Hx3-192, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 64.6 | 62.3 | 49.6 | — | 96.4 | 95.6 | 89.5 | 10.5 | |
| ASFD-D1Single Path=R18-FA-Hx3-128, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 64.1 | 61.5 | 48.9 | — | 96.1 | 95.2 | 88.8 | 5.7 | |
| ASFD-D0Single Path=R18-FA-Hx1-64, Resolution=VGA, Hardware=Nvidia V100 GPU2022.01 | 63.7 | 61.1 | 48.3 | — | 95.7 | 94.8 | 88 | 3.1 | |
| Blacked outFace detector=DSFD [14], Anonymization method=Blacked out2019.09 | 24.9 | 36.3 | 54.8 | — | — | — | — | — | |
| CSR-A-thr (baseline)Supervision=Baseline2019.06 | — | — | — | 41.9 | — | — | — | — | |
| FacenessSupervision=Full supervision2019.06 | — | — | — | 53.4 | — | — | — | — | |
| img2poseTraining Dataset=WIDER2024.07 | — | — | — | — | 86.5 | 82.9 | 61.3 | — | |
| LSC-CNN (Actual GT)Supervision=Actual ground truths2019.06 | — | — | — | 70.1 | — | — | — | — | |
| LSC-CNN (Pseudo GT)Supervision=Pseudo ground truths2019.06 | — | — | — | 62.1 | — | — | — | — | |
| PSDNNSupervision=Pseudo ground truths2019.06 | — | — | — | 60.5 | — | — | — | — | |
| RetinaFaceTraining Dataset=WIDER2024.07 | — | — | — | — | 94.6 | 93 | 80.4 | — | |
| SSHSupervision=Full supervision2019.06 | — | — | — | 92.1 | — | — | — | — | |
| TinyFaceSupervision=Full supervision2019.06 | — | — | — | 91 | — | — | — | — | |
| Two Stage CNNSupervision=Full supervision2019.06 | — | — | — | 61.4 | — | — | — | — | |
| VGGHeadsTraining Dataset=VGGHeads2024.07 | — | — | — | — | 56.3 | 51 | 29.2 | — | |
| VGGHeadsTraining Dataset=VGGHeads + WIDER2024.07 | — | — | — | — | 92.6 | 88.9 | 70.3 | — |