DeepLabV3 Weights
Pre-release
Pre-release
DeepLabV3
Rethinking Atrous Convolution for Semantic Image Segmentation
Results
Cityscapes
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|---|---|---|---|---|---|---|---|---|
| resnet18 | 512x1024 | 89 | 61.41 | 16G | 498MB | 21G |
SBD
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|---|---|---|---|---|---|---|---|---|
| resnet18 | 500x500 | 93 | 61.24 | 16G | 273MB | 10G |