Research map / Continual learning / Class-incremental
Continual Semantic Segmentation With Automatic Memory Sample Selection
Read the paperCheck 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 automatic RL-driven memory selection and gradient-based sample enhancement instead of random memory filling (ablation) | Continual semantic segmentation · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 63.15 -> 71.94 | Helped |
| Add gradient-based enhancement to selected memory samples (ablation) | RL-selected replay memory · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 70.02 -> 71.94 | Helped |
| Use learned sample selection rather than random memory filling (ablation) | Continual semantic segmentation · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 63.15 -> 70.02 | Helped |
| Add sample diversity to the state representation (ablation) | RL memory-selection policy · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 64.39 -> 71.94 | Helped |
| Add class accuracy to the state representation (ablation) | RL memory-selection policy · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 68.30 -> 71.94 | Helped |
| Add class forgetfulness to the state representation (ablation) | RL memory-selection policy · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 70.06 -> 71.94 | Helped |
| Add both class accuracy and forgetfulness to the state (ablation) | RL memory-selection policy · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 66.68 -> 71.94 | Helped |
| Use graph-based rather than prototype-based sample diversity (ablation) | RL memory-selection policy · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 69.83 -> 71.94 | Helped |
| Increase memory length from 50 to 300 (ablation) | Continual semantic segmentation · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 63.56 -> 75.33 | Helped |
| Use all current-class images instead of a randomly sampled 10% support set (ablation) | State representation computation · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 71.94 -> 72.25 | No clear effect |
| Use RL-learned selection rather than hand-crafted selection with the same factors | Memory sample selection · Pascal-VOC 2012 15-1 (6 stages) | mIoU (all): 66.82 -> 70.02 | Helped |
Closest papers by meaning
- Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · similarity 0.91)
- TIKP: Text-to-Image Knowledge Preservation for Continual Semantic Segmentation (AAAI 2024 · similarity 0.89)
- Adaptive Prototype Replay for Class Incremental Semantic Segmentation (AAAI 2025 · similarity 0.88)
- Towards Continual Universal Segmentation (CVPR 2025 · similarity 0.87)
- Rethinking Query-based Transformer for Continual Image Segmentation (CVPR 2025 · similarity 0.87)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
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.