Research map / Continual learning / Class-incremental
ALIFE: Adaptive Logit Regularizer and Feature Replay for Incremental Semantic Segmentation
Youngmin Oh, Donghyeon Baek, Bumsub Ham
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 feature replay with rotation matrices (ALIFE-M vs ALIFE) | ALIFE · ADE20K 100-50(1) | hIoU: 29.83 -> 30.28 | Improved |
| add feature replay with rotation matrices | ALIFE · VOC 16-5(5) | hIoU: 45.29 -> 48.89 | Improved |
| replace CCE/CKD (MiB) with ALI | MiB · ADE20K 100-50(1) | hIoU: 24.12 -> 29.83 | Improved |
| add ALI to CE loss (ablation) | ALIFE · VOC 16-5(1) | hIoU: 9.74 -> 60.02 | Improved |
| add KD on labeled regions to CE+ALI (ablation) | ALIFE · VOC 16-5(1) | hIoU: 60.02 -> 62.50 | Improved |
| apply KD on unlabeled regions too (ablation) | ALIFE · VOC 16-5(1) | hIoU: 60.02 -> 59.97 | Hurt |
| apply KD on all regions instead of labeled only (ablation) | ALIFE · VOC 16-5(1) | hIoU: 62.50 -> 61.12 | Hurt |
| pseudo-label unlabeled regions in classifier fine-tuning (FL) (ablation) | ALIFE step 3 · VOC 16-5(1) | hIoU: 60.26 -> 61.29 | Improved |
| add 1K memorized features per previous category (ablation) | ALIFE step 3 · VOC 16-5(1) | hIoU: 61.29 -> 64.17 | Improved |
| add ALI in step 3 without memory (ablation) | ALIFE step 3 · VOC 16-5(1) | hIoU: 61.29 -> 62.78 | Improved |
| ALIFE-M vs SSUL-M | ALIFE-M · VOC 16-5(5) | hIoU: 48.89 vs 60.30 | Hurt |
Later papers that compared with this method
1 later accepted paper reports ALIFE as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Preparing the Future for Continual Semantic Segmentation (ICCV 2023 · better on 4 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SSUL — SSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental Learning (NeurIPS 2021) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Modeling the Background for Incremental Learning in Semantic Segmentation (CVPR 2020 · similarity 0.89)
- A Survey on Continual Semantic Segmentation: Theory, Challenge, Method and Application (TPAMI 2024 · similarity 0.89)
- Mining Unseen Classes via Regional Objectness: A Simple Baseline for Incremental Segmentation (NeurIPS 2022 · similarity 0.89)
- Incremental Learning in Semantic Segmentation From Image Labels (CVPR 2022 · similarity 0.89)
- Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · similarity 0.88)
- TIKP: Text-to-Image Knowledge Preservation for Continual Semantic Segmentation (AAAI 2024 · 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.
