Opt: Optimizer during training
Iter: Training iterations
Aux: Apply auxiliary loss during training
| Methods | Backbone | Opt | Iter | Aux | SS mIOU(%) | MF mIOU(%) | ckpts | Logs |
|---|---|---|---|---|---|---|---|---|
| FCN | ResNet-50 (D8) | SGD | 30K | 47.72 | - | Baidu Drive | tb.dev | |
| FCN + CAR | ResNet-50 (D8) | SGD | 30K | 48.40 | - | Baidu Drive | tb.dev | |
| Self-Attention | ResNet-50 (D8) | SGD | 30K | 48.32 | - | Baidu Drive | tb.dev | |
| Self-Attention + CAR | ResNet-50 (D8) | SGD | 30K | 50.50 | - | Baidu Drive | tb.dev | |
| CAA | ConvNeXt-L (JPU) | SGD | 30K | 60.48 | 61.80 | Baidu Drive | tb.dev | |
| CAA + CAR | ConvNeXt-L (JPU) | SGD | 30K | 61.40 | 62.69 | Baidu Drive | tb.dev | |
| CAA + CAR | ConvNeXt-L (JPU) | Adam | 40K | ✓ | 62.97 | 64.12 | Baidu Drive | tb.dev |
More ckpts will be uploaded later (I am busy atm), create an issue if you need them now.