Semantic Segmentation on PASCAL VOC augmented 2012 (1% labeled)
41.41mIoUCompL (DenseCL)
Evaluation results
| Method | Links | |
|---|---|---|
| CompL (DenseCL)Training Strategy=Joint Training, Auxiliary Task=DenseCL, Backbone=ResNet-1012022.10 | 41.41 | |
| Semseg BaselineTraining Strategy=Supervised Baseline, Backbone=ResNet-1012022.10 | 38.87 | |
| DenseCL → Semseg + DenseCLTraining Strategy=Transfer + Joint Training (CompL), Auxiliary Task=DenseCL, Backbone=ResNet-262022.10 | 36.32 | |
| CompL (DenseCL)Training Strategy=Joint Training, Auxiliary Task=DenseCL, Backbone=ResNet-502022.10 | 36.15 | |
| DenseCL → SemsegTraining Strategy=Transfer Learning, Pre-training Task=DenseCL, Backbone=ResNet-262022.10 | 34.89 | |
| Semseg BaselineTraining Strategy=Supervised Baseline, Backbone=ResNet-502022.10 | 34.87 | |
| Semseg + DenseCLTraining Strategy=Joint Training (CompL), Auxiliary Task=DenseCL, Backbone=ResNet-262022.10 | 33.51 | |
| Semseg + MoCoTraining Strategy=Joint Training (CompL), Auxiliary Task=MoCo, Backbone=ResNet-262022.10 | 32.9 | |
| MoCo → SemsegTraining Strategy=Transfer Learning, Pre-training Task=MoCo, Backbone=ResNet-262022.10 | 31.55 | |
| SemsegTraining Strategy=Supervised Baseline, Backbone=ResNet-262022.10 | 30.82 | |
| Semseg + RotTraining Strategy=Joint Training (CompL), Auxiliary Task=Rotation, Backbone=ResNet-262022.10 | 28.75 | |
| Rot → SemsegTraining Strategy=Transfer Learning, Pre-training Task=Rotation, Backbone=ResNet-262022.10 | 10.35 |