Research map / Continual learning / Class-incremental
Learning to Segment the Tail
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 |
|---|---|---|---|
| Apply class-incremental LST instead of the baseline | ResNeXt-101-32x8d-FPN Mask R-CNN · LVIS val, classes with <10 instances | AP: 3.5 -> 14.4 | Helped |
| Apply class-incremental LST instead of the baseline | ResNeXt-101-32x8d-FPN Mask R-CNN · LVIS val | AP: 23.0 -> 26.3 | Helped |
| Add the meta weight generator (ablation) | LST · LVIS val, classes with <5 instances | AP: 5.3% vs 8.0% | Helped |
| Add the meta weight generator (ablation) | LST · LVIS val, classes with <10 instances | AP: 11.7 -> 11.2 | Mixed |
| Add knowledge distillation between adjacent phases (ablation) | LST · LVIS, all classes | AP: 21.6% -> 22.8% | Helped |
| Use balanced replay instead of replaying all old-class data (ablation) | LST · LVIS incremental phases, old and new classes | instance segmentation performance | Helped |
| Increase base classes from b=270 to b=590 (ablation) | LST · LVIS val | AP: 22.8 -> 21.2 | Hurt |
| Use a single large incremental phase (phase_size=960) rather than 6 phases of size 160 (ablation) | LST with b=270 · LVIS val | AP: 22.8 -> 21.8 | Hurt |
| Use moderate phase sizes of 320 or 480 rather than 160 (ablation) | LST with b=270 · LVIS val | AP: 22.8 -> 22.9 | No clear effect |
Closest papers by meaning
- Rethinking Class-Balanced Methods for Long-Tailed Visual Recognition From a Domain Adaptation Perspective (CVPR 2020 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.86)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.86)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.86)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.86)
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.85)
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.