Portrait Matting on RWP (test)
16.98SAD (Trimap)DIM*
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DIM*Backbone=VGG16, Trimap usage=ground truth trimaps, Training data=P3M-10k face-blurred training set2022.03 | 16.98 | — | — | — | 0.0508 | 0.1052 | — | — | 46.47 | 16.93 | |
| P3M(S)+FCPBackbone=Swin-T, Post-processing=FCP, Training data=P3M-10k face-blurred training set2022.03 | 21.32 | 31.11 | 0.0149 | 0.0231 | 0.0728 | 0.1325 | 3.108 | 6.682 | 56.83 | 31 | |
| P3M(S)Backbone=Swin-T, Training data=P3M-10k face-blurred training set2022.03 | 21.38 | 32.71 | 0.0157 | 0.0236 | 0.0744 | 0.1326 | 3.999 | 7.328 | 57.48 | 32.48 | |
| P3M(V)Backbone=ViTAE-S, Training data=P3M-10k face-blurred training set2022.03 | 21.43 | 29 | 0.0151 | 0.0229 | 0.076 | 0.1344 | 1.75 | 5.822 | 56.66 | 28.77 | |
| P3M(R)Backbone=ResNet-34, Training data=P3M-10k face-blurred training set2022.03 | 22.25 | 36.47 | 0.0183 | 0.027 | 0.0803 | 0.1394 | 5.895 | 8.325 | 60.82 | 36.36 | |
| MODNetBackbone=MobileNet V2, Training data=P3M-10k face-blurred training set2022.03 | 24.84 | 42.63 | 0.0227 | 0.0338 | 0.0989 | 0.1611 | 5.84 | 11.95 | 66.26 | 39.63 | |
| SHMBackbone=PSPNet-50+VGG16, Training data=P3M-10k face-blurred training set2022.03 | 25.01 | 48.95 | 0.0283 | 0.0375 | 0.1013 | 0.1629 | 6.19 | 17.747 | 68.52 | 48.74 | |
| GFMBackbone=ResNet-34, Training data=P3M-10k face-blurred training set2022.03 | 25.49 | 37.93 | 0.022 | 0.0316 | 0.098 | 0.1672 | 4.146 | 8.3 | 67.56 | 37.48 | |
| HATTBackbone=ResNeXt, Training data=P3M-10k face-blurred training set2022.03 | 28.99 | 53.45 | 0.0234 | 0.0376 | 0.117 | 0.1856 | 8.332 | 16.125 | 78.9 | 46.71 | |
| LFBackbone=DenseNet-201, Training data=P3M-10k face-blurred training set2022.03 | 31.28 | 71.18 | 0.0423 | 0.0556 | 0.1281 | 0.1972 | 20.027 | 19.879 | 71.64 | 70.54 |