Research map / Label-efficient and robust learning / Semi-supervised learning
Improving Semi-Supervised and Domain-Adaptive Semantic Segmentation with Self-Supervised Depth Estimation
Lukas Hoyer, Dengxin Dai, Qin Wang, Yuhua Chen, Luc Van Gool
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DepthMix instead of ClassMix (ablation) | semi-supervised segmentation · Cityscapes (372 labels) | mIoU (%): 63.16 -> 64.14 | Improved |
| add DepthMix instead of ClassMix (ablation) | semi-supervised segmentation · Cityscapes (2975 labels) | mIoU (%): 69.60 -> 69.83 | Improved |
| add mean-teacher pseudo-labels (ablation) | baseline · Cityscapes (372 labels) | mIoU (%): 59.14 -> 62.39 | Improved |
| add DepthMix on top of pseudo-labels (ablation) | pseudo-label baseline · Cityscapes (372 labels) | mIoU (%): 62.39 -> 64.14 | Improved |
| add SDE transfer learning of encoder and decoder (ablation) | baseline · Cityscapes (372 labels) | mIoU (%): 59.14 -> 60.46 | Improved |
| add ImageNet feature distance loss to SDE pretraining (ablation) | SDE transfer · Cityscapes (372 labels) | mIoU (%): 60.46 -> 60.80 | Improved |
| add multi-task learning to transfer (ablation) | SDE transfer + F · Cityscapes (372 labels) | mIoU (%): 60.80 -> 61.25 | Improved |
| add diversity plus uncertainty sampling instead of diversity only (ablation) | automatic data selection · Cityscapes (100 labels) | mIoU (%): 53.00 -> 54.37 | Improved |
| diversity vs uncertainty sampling at few labels (ablation) | automatic data selection · Cityscapes (100 labels) | mIoU (%): 53.00 vs 51.75 | Mixed |
| uncertainty sampling beats diversity with more labels (ablation) | automatic data selection · Cityscapes (744 labels) | mIoU (%): 66.76 vs 66.37 | Mixed |
| add depth-based DS+US selection instead of entropy active learning | random baseline · Cityscapes (372 labels) | mIoU (%): 63.51 -> 64.25 | Improved |
| add data selection to DepthMix+MTL (ablation) | DepthMix+MTL · Cityscapes (372 labels) | mIoU (%): 66.66 -> 68.01 | Improved |
Closest papers by meaning
- Three Ways To Improve Semantic Segmentation With Self-Supervised Depth Estimation (CVPR 2021 · similarity 0.96)
- Domain Adaptive Semantic Segmentation With Self-Supervised Depth Estimation (ICCV 2021 · similarity 0.94)
- Improving Semantic Segmentation via Efficient Self-Training (TPAMI 2021 · similarity 0.92)
- Universal Semi-Supervised Semantic Segmentation (ICCV 2019 · similarity 0.91)
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · similarity 0.90)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
