Table 6 Performance comparison between baseline and ECDF-enhanced architectures, demonstrating the effectiveness of replacing conventional feature extraction modules with our proposed ECDF block.
From: Lightweight Unet with depthwise separable convolution for skin lesion segmentation
Model | Use block | Params (M) | FLOPS (G) | Dice (ISIC2018) |
|---|---|---|---|---|
SegNet | Original ECDF | 29. 44 0.620 | 30.75 6.980 | 0. 8427 0. 8496 |
TransUnet | Original ECDF | 66.82 38.40 | 25.00 13.32 | 0. 8256 0. 8241 |
