Semantic Segmentation on Cityscapes (val) (mIoU, GMACS, FPS)
80.6mIoUStochastic Sampling
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Stochastic Sampling2020.11 | 80.6 | 373.2 | — | — | |
| SwiftNet-RN50 (baseline, our impl.)Backbone=ResNet-50, Evaluation Device=1080 Ti2020.11 | 78 | 206.3 | 16.3 | 16.3 | |
| SegBlocks-RN50Backbone=ResNet-50, Hyperparameter τ=0.4, Evaluation Device=1080 Ti2020.11 | 77.5 | 127.2 | 23.3 | 23.3 | |
| SwiftNet-EffL1 (baseline, our impl.)Backbone=EfficientNet-L1, Evaluation Device=1080 Ti2020.11 | 76.6 | 24.9 | 17.4 | 17.4 | |
| Huang et al.Evaluation Device=1080 Ti2020.11 | 76.4 | — | 1.8 | 1.8 | |
| SegBlocks-RN18Backbone=ResNet-18, Hyperparameter τ=0.4, Evaluation Device=1080 Ti2020.11 | 76.3 | 60.5 | 48.6 | 48.6 | |
| SegBlocks-EffL1Backbone=EfficientNet-L1, Hyperparameter τ=0.4, Evaluation Device=1080 Ti2020.11 | 76.3 | 15.6 | 30.4 | 30.4 | |
| SegBlocks-RN50Backbone=ResNet-50, Hyperparameter τ=0.2, Evaluation Device=1080 Ti2020.11 | 76.2 | 88.5 | 30 | 30 | |
| SwiftNet-RN18 (baseline, our impl.)Backbone=ResNet-18, Evaluation Device=1080 Ti2020.11 | 76.2 | 104.1 | 38.4 | 38.4 | |
| HyperSeg-MEvaluation Device=1080 Ti2020.11 | 76.2 | 7.5 | 36.9 | 36.9 | |
| SegBlocks-RN18Backbone=ResNet-18, Hyperparameter τ=0.2, Evaluation Device=1080 Ti2020.11 | 75.9 | 43.5 | 57.8 | 57.8 | |
| BiSeNetV2-RN18Backbone=ResNet-18, Evaluation Device=1080 Ti2020.11 | 75.8 | 118.5 | 47.3 | 47.3 | |
| SwiftNet-RN18Backbone=ResNet-18, Evaluation Device=1080 Ti2020.11 | 75.6 | 104 | 39.9 | 39.9 | |
| SegBlocks-EffL1Backbone=EfficientNet-L1, Hyperparameter τ=0.2, Evaluation Device=1080 Ti2020.11 | 75.3 | 11.4 | 35.6 | 35.6 | |
| Patch Proposal NetworkEvaluation Device=1080 Ti2020.11 | 75.2 | — | 24 | 24 | |
| BiSeNet-RN18Backbone=ResNet-18, Evaluation Device=Titan Xp2020.11 | 74.8 | 67 | 65.5 | 58.5 | |
| Wu et al.Evaluation Device=980 Ti2020.11 | 72.9 | — | 15 | 33 | |
| ERFNetEvaluation Device=Titan X2020.11 | 72.7 | 27.7 | 11.2 | 18.4 | |
| DABNetEvaluation Device=1080 Ti2020.11 | 70.1 | — | 27.7 | 27.7 | |
| ICNetEvaluation Device=Titan X2020.11 | 69.5 | 30 | 30.3 | 49.7 | |
| ESPNetV22020.11 | 66.1 | 2.7 | — | — | |
| Learning Downsampling2020.11 | 65 | 34 | — | — |