Research map / Detection and segmentation / Weakly supervised detection and segmentation
Incremental Learning in Semantic Segmentation From Image Labels
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 |
|---|---|---|---|
| Generate pseudo-supervision online using a localizer and the old segmentation model's localization prior rather than offline pseudo-labels | WILSON · Pascal VOC 10-10 overlap | mIoU: +0.7% overall mIoU | Helped |
| Add localization supervision from the previous segmentation model using LOC loss rather than no prior (ablation) | WILSON localizer · Pascal VOC 10-10 disjoint, new classes | mIoU: -4.4% | Helped |
| Add localization supervision from the previous segmentation model using LOC loss rather than no prior (ablation) | WILSON localizer · Pascal VOC 10-10 overlap, new classes | mIoU: -5.1 | Helped |
| Use a learned localization prior instead of directly concatenating the old model's output as a fixed prior (ablation) | WILSON localizer · Pascal VOC 10-10 disjoint | mIoU: -4.0% | Helped |
| Use a learned localization prior instead of directly concatenating the old model's output as a fixed prior (ablation) | WILSON localizer · Pascal VOC 10-10 overlap | mIoU: -4.3% | Helped |
| Use independent logistic LOC loss rather than softmax cross-entropy to match old-model outputs (ablation) | WILSON localizer · Pascal VOC 10-10 disjoint | mIoU: -6.3% | Helped |
| Use independent logistic LOC loss rather than softmax cross-entropy to match old-model outputs (ablation) | WILSON localizer · Pascal VOC 10-10 overlap | mIoU: -5.8% | Helped |
| Blend hard and soft pseudo-labels rather than use hard labels alone (ablation) | WILSON segmentation model · Pascal VOC 10-10 | mIoU | Helped |
| Vary alpha from 0 to 0.7 when blending hard and soft pseudo-labels (ablation) | WILSON segmentation model · Pascal VOC 10-10 disjoint and overlap | mIoU: less than 0.5% | No clear effect |
| Provide image-level supervision for old classes as well as new classes (ablation) | WILSON · Pascal VOC 15-5 and 10-10 | mIoU: 2% | Helped |
| Provide image-level supervision for old classes as well as new classes (ablation) | WILSON · Pascal VOC 15-5 disjoint and overlap | mIoU: 1.2% and 0.1% | Helped |
| Provide image-level supervision for old classes as well as new classes (ablation) | WILSON · Pascal VOC 10-10 disjoint and overlap | mIoU: 1.8% and 1.3% | Helped |
Closest papers by meaning
- Foundation Model Drives Weakly Incremental Learning for Semantic Segmentation (CVPR 2023 · similarity 0.93)
- Single-Stage Semantic Segmentation From Image Labels (CVPR 2020 · similarity 0.91)
- Modeling the Background for Incremental Learning in Semantic Segmentation (CVPR 2020 · similarity 0.91)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.89)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.88)
- Separate and Conquer: Decoupling Co-occurrence via Decomposition and Representation for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.88)
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.