Table 2.
Comparison of various upsampling schemes for semantic segmentation. The results are presented both on the individual MosMeddata and Jun Ma datasets as well as on the combined set. The same test set partition is used to test the model.
| S. No. | Method | Jun Ma dataset [30] | MosMedData [31] | Combined dataset | |||
|---|---|---|---|---|---|---|---|
| DSC | IoU | DSC | IoU | DSC | IoU | ||
| 1 | Bilinear upsampling | 70.49 | 58.80 | 63.54 | 52.17 | 66.27 | 53.15 |
| 2 | Sub-pixel shuffling dense upsample [20] | 73.52 | 59.69 | 67.83 | 55.47 | 69.12 | 57.77 |
| 3 | Global Attention Upsample [18] | 71.63 | 57.56 | 70.97 | 58.38 | 71.35 | 60.01 |
| 4 | Attention-guided dense-upsampling [19] | 78.17 | 66.32 | 73.02 | 58.97 | 74.86 | 61.88 |
| 5 | Data-dependent Upsampling [31] | 77.31 | 64.80 | 74.95 | 58.62 | 75.92 | 63.12 |
| 6 | Proposed CCAF upsampler CNN | 80.43 | 69.87 | 75.19 | 65.30 | 77.67 | 65.79 |