Semantic Segmentation on RIM-One target domain
95.9Dice (Disc)L_sup
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| L_suptarget_data_percentage=50%2021.07 | 95.9 | 83.4 | |
| Polyformerloss_components=L_sup + L_adv + K2021.07 | 91.3 | 75.8 | |
| Polyformerloss_components=L_sup + K2021.07 | 90 | 75.3 | |
| DA-ADVfine-tuning_strategy=tune whole model2021.07 | 88.5 | 72.5 | |
| ADDAloss_components=L_sup + L_adv2021.07 | 87.4 | 72.6 | |
| DA-ADVfine-tuning_strategy=tune last two layers2021.07 | 87.2 | 73 | |
| L_suptraining_data=target-only2021.07 | 87.1 | 66.5 | |
| CellSegSSDAloss_components=L_sup + L_adv (mask) + L_recon2021.07 | 86.9 | 75.6 | |
| RevGradloss_components=L_sup + L_adv2021.07 | 86 | 73.2 | |
| L_suptraining_data=source + target2021.07 | 83.1 | 71.5 | |
| U-Nettraining_data=source-only2021.07 | 81.9 | 70.8 | |
| CycleGAN + L_suptraining_data=source2021.07 | 74.7 | 69 |