Instance Segmentation on ADE20K 80 x 80 (test)
56mIoUFSNet+HRNet
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FSNet+HRNetDownsampling Strategy=FSNet, Backbone=HRNet, Resolution=80 x 802025.03 | 56 | 56 | |
| FSNet+DeepLabDownsampling Strategy=FSNet, Backbone=DeepLabV3, Resolution=80 x 802025.03 | 55 | 56 | |
| FSNet+SegFormer-B5Downsampling Strategy=FSNet, Backbone=SegFormer-B5, Resolution=80 x 802025.03 | 55 | 57 | |
| FSNet+SegFormer-B4Downsampling Strategy=FSNet, Backbone=SegFormer-B4, Resolution=80 x 802025.03 | 54 | 55 | |
| Avg+HRNetDownsampling Strategy=Uniform Subsampling, Backbone=HRNet, Resolution=80 x 802025.03 | 43 | 44 | |
| Avg+SegFormer-B5Downsampling Strategy=Uniform Subsampling, Backbone=SegFormer-B5, Resolution=80 x 802025.03 | 41 | 42 | |
| LTDDownsampling Strategy=Learn-to-Downsample, Resolution=80 x 802025.03 | 41 | 41 | |
| Avg+DeepLabDownsampling Strategy=Uniform Subsampling, Backbone=DeepLabV3, Resolution=80 x 802025.03 | 39 | 41 | |
| Avg+SegFormer-B4Downsampling Strategy=Uniform Subsampling, Backbone=SegFormer-B4, Resolution=80 x 802025.03 | 37 | 39 |