Table 1.
The contributions of CNN networks.
Fig 1.
The overall architecture of TC-Net.
The image to be processed and input it to the dual coding channel at the same time, then fuse at the bottom of the coding to decode and output the split image.
Fig 2.
The architecture of feature fusion block.
Fig 3.
The architecture of CNN and transformer fusion blocks.
Table 2.
The introduction of the public datasets.
Table 3.
The ablation experiment based on ISIC2018.
Fig 4.
Visual analysis of ablation experiment on ISBI2016.
(a) Original. (b) Gray. (c) GT. (d) CE-Net. (e) Swin-Unet. (f) Double coding+add. (g) TC-Net.
Fig 5.
Visual analysis of ablation experiment on ISBI2016.
(a) Original. (b) Gray. (c) GT. (d) CE-Net. (e) Swin-Unet. (f) Double coding+add. (g) TC-Net.
Table 4.
Comparative experiments based on ISIC2018 dataset.
Table 5.
Comparative experiments based on ISBI2017 dataset.
Table 6.
Comparative experiments based on ISBI2016 dataset.
Table 7.
Comparative experiences with state-of-the-art methods on fused networks.
Fig 6.
The example on ISIC2018 dataset, (a) original image; (b) Gray image; (c) GT label image; (d) Segmentation image of CE-Net; (e) Segmentation image of Swin-UNet; (f) Segmentation image of TC-Net.
Fig 7.
The example on ISIC2018 dataset, (a) original image; (b) Gray image; (c) GT label image; (d) Segmentation image of CE-Net; (e) Segmentation image of Swin-UNet; (f) Segmentation image of TC-Net.
Fig 8.
The detail analysis on ISBI2017 dataset, (a) original image; (b) GT label image; (c) Segmentation image of CE-Net; (d) Segmentation image of Swin-UNet; (e) Segmentation image of TC-Net.The red box indicates the segmentation edge information at the same position.