Research Article
[Retracted] DARSegNet: A Real-Time Semantic Segmentation Method Based on Dual Attention Fusion Module and Encoder-Decoder Network
Table 3
Comparison on cityscapes test set.
| Method | Pretraining | GPU | MIoU | Speed (FPS) | Params |
| SegNet | ImageNet | TitanX | 56.1 | 15 | 29.5 M | ENet | No | TitanX | 58.3 | 31 | 0.36 M | ESPNet | No | TitanX | 60.3 | 113 | 0.36 M | CGNet | No | 1080Ti | 64.8 | 17 | 0.50 M | ContextNet | ImageNet | 1080Ti | 66.1 | 18 | 0.85 M | EDANet | No | 1080Ti | 67.3 | 81 | 0.68 M | ERFNet | No | 1080Ti | 68.0 | 42 | 2.10 M | BiseNet | ImageNet | 1080Ti | 68.4 | 106 | 5.80 M | ICNet | ImageNet | TitanX M | 69.5 | 30 | 7.80 M | DABNet | No | 1080Ti | 70.1 | 28 | 0.80 M | BiSeNet V2 | No | 1080Ti | 73.4 | 156 | 49M | BiSeNet V2-L | No | 1080Ti | 75.8 | 47.3 | NULL | DARSegNet (ours) | No | 1080Ti | 75.49 | 69.3 | 4.15 M | DARSegNet (ours) | ImageNet | 1080Ti | 75.53 | 69.5 | 4.15 M |
|
|