Weakly supervised semantic segmentation on VOC 2012 (val)
70mIoUPMM
Evaluation Results
| Method | Links | |
|---|---|---|
| PMMExtra Info=None, Backbone=Res2Net2021.08 | 70 | |
| LIIDExtra Info=Segment-based object proposals (SOP), Backbone=VGG2021.08 | 69.4 | |
| PMMExtra Info=None, Backbone=ResNet2021.08 | 68.5 | |
| Li et al.Extra Info=External saliency models (S), Backbone=ResNet2021.08 | 68.2 | |
| ICDExtra Info=External saliency models (S), Backbone=ResNet2021.08 | 67.8 | |
| SGANExtra Info=External saliency models (S), Backbone=ResNet2021.08 | 67.1 | |
| PMMExtra Info=None, Backbone=ScaleNet2021.08 | 67.1 | |
| LIIDExtra Info=Segment-based object proposals (SOP), Backbone=ResNet2021.08 | 66.5 | |
| SDIExtra Info=D+BSDS, Backbone=ResNet2021.08 | 65.7 | |
| OAA+Extra Info=External saliency models (S), Backbone=ResNet2021.08 | 65.2 | |
| FickleNetExtra Info=External saliency models (S), Backbone=ResNet2021.08 | 64.9 | |
| BoxSupExtra Info=Supervision of detection (D), Backbone=VGG2021.08 | 62 | |
| DSRGExtra Info=MSRA-B, Backbone=ResNet2021.08 | 61.4 | |
| PRMComments=Object segment proposals, Negligible training costs=true2018.04 | 53.4 | |
| CombiningComments=CRF as RNN, Negligible training costs=false2018.04 | 52.8 | |
| Check maskComments=CRF & Human in the loop, Negligible training costs=false2018.04 | 51.5 | |
| SECComments=CRF as boundary loss, Negligible training costs=false2018.04 | 50.7 | |
| WILDCATComments=CRF post-processing, Negligible training costs=true2018.04 | 43.7 | |
| MIL+ILP+SP-segComments=Object segment proposals, Negligible training costs=true2018.04 | 42 |