Research map / Label-efficient and robust learning / Semi-supervised learning
Three Ways To Improve Semantic Segmentation With Self-Supervised Depth Estimation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use SDE transfer learning with multi-task learning and ImageNet feature-distance preservation (ablation) | Cityscapes semantic segmentation · Cityscapes val, 372 and 2975 labels | mIoU (%): +2.10 (+1.99) | Helped |
| Add ImageNet feature-distance loss during SDE pretraining (ablation) | SDE transfer learning · Cityscapes val, 372 and 2975 labels | mIoU (%): 60.46 -> 60.80; 69.00 -> 69.47 | Helped |
| Add multi-task SDE learning (ablation) | SDE transfer with ImageNet feature distance · Cityscapes val, 372 and 2975 labels | mIoU (%): 60.80 -> 61.25; 69.47 -> 69.76 | Helped |
| Add pseudo-labeling with ClassMix (ablation) | Cityscapes semantic segmentation · Cityscapes val, 372 and 2975 labels | mIoU (%): +4.02 (+1.83) | Helped |
| Use geometry-aware DepthMix instead of class-based mixing (ablation) | Pseudo-labeled semantic segmentation · Cityscapes val, 372 and 2975 labels | mIoU (%): 63.16 -> 64.14; 69.60 -> 69.83 | Helped |
| Add automatic data selection (ablation) | Cityscapes semantic segmentation · Cityscapes val, 372 labels | mIoU (%): +5.11 | Helped |
| Combine SDE multi-task learning, DepthMix, and automatic data selection (ablation) | Cityscapes semantic segmentation · Cityscapes val, 372 labels | mIoU (%): +8.87 | Helped |
| Use combined depth-feature diversity and depth-student uncertainty for selection | Cityscapes data selection · 100, 372, and 744 selected labels | mIoU (%): 54.37; 64.25; 66.94 | Helped |
| Add unlabeled video sequences to ClassMix | ClassMix · Cityscapes val, full (2975 labels) | mIoU (%): 68.23 | No clear effect |
| Transfer SDE features and add DepthMix plus data selection | CamVid semantic segmentation · CamVid test, 50 labels | mIoU (%): +9.07 | Helped |
Closest papers by meaning
- Semantically-Guided Representation Learning for Self-Supervised Monocular Depth (ICLR 2020 · similarity 0.91)
- Semi-Supervised Semantic Image Segmentation With Self-Correcting Networks (CVPR 2020 · similarity 0.91)
- Learning from Future: A Novel Self-Training Framework for Semantic Segmentation (NeurIPS 2022 · similarity 0.91)
- ST++: Make Self-Training Work Better for Semi-Supervised Semantic Segmentation (CVPR 2022 · similarity 0.90)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.90)
- AllSpark: Reborn Labeled Features from Unlabeled in Transformer for Semi-Supervised Semantic Segmentation (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.