Results
Cityscapes
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 512x1024 | 96 | 62.09 | | 11M | 468MB | 19G | |
sfnet-lite
Semantic Flow Network Lite
Results
Cityscapes
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 512x1024 | 97 | 64.92 | | 12M | 632MB | 30G | |
| resnet101 | 512x1024 | 51 | 71.38 | | 44M | 3.0GB | 93G | |
SBD
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 500x500 | 99 | 64.84 | | 12M | 329MB | 14G | |
sfnet
Semantic Flow Network
[Semantic Flow for Fast and Accurate Scene Parsing](https://arxiv.org/pdf/2002.10120)
Results
Cityscapes
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 512x1024 | 87 | 66.78 | | 13M | 750MB | 46G | |
| resnet101 | 512x1024 | 97 | 72.01 | | 45M| 3.1GB | 110G | |
SBD
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 500x500 | 57 | 64.31 | | 13M | 387MB | 22G | |
xception
Xception
[Xception: Deep Learning with Depthwise Separable Convolutions](https://arxiv.org/abs/1610.02357)
Weights are downloaded and converted from keras <https://github.com/fchollet/deep-learning-models/releases/tag/v0.4>
Implementation and results can be found in <https://github.com/keras-team/keras-applications/blob/master/keras_applications/xception.py>
enet
ENet
[ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation](https://arxiv.org/abs/1606.02147)
Results
Cityscapes
| Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|
| 512x1024 | 99 | 47.48 | | 346K | 1.3GB | 5.39G | |
SBD
| Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|
| 500x500 | | | | 347K | 612MB | 2.59G | |
bisenet
Bilateral Segmentation Network
[BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation](https://arxiv.org/abs/1808.00897)
Results
Cityscapes
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 512x1024 | 95 | 62.63 | | 14M | 846MB | 32G | |
| resnet50 | 512x1024 | 98 | 68.24 | | 29M | 4.6GB | 241G | |
| xception | 512x1024 | 100 | 69.21 | | 26M | 3.1GB | 63G | |
SBD
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|
| resnet18 | 500x500 | 82 | 62.56 | 63.68 | 14M | 438MB | 16G | 5.4ms |
pspnet
Pyramid Scene Parsing Network
[Pyramid Scene Parsing Network](https://arxiv.org/abs/1612.01105)
Results
Cityscapes
| Backbone | Train size | Epoch | mIoU | mIoU(tta) | Params | Memory | MACs | Time |
|:-|:-|:-|:-|:-|:-|:-|:-|:-|