Research map / Continual learning / Class-incremental
Overcoming Catastrophic Forgetting in Incremental Object Detection via Elastic Response Distillation
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 ERD instead of fine-tuning on new data only | GFLV1 · COCO minival, 40 classes + 40 classes | AP: 17.8 -> 36.9 | Helped |
| use ERD rather than SID | GFLV1 · COCO minival, 40 classes + 40 classes | AP: 34.0 -> 36.9 | Helped |
| use ERD rather than LwF | GFLV1 · COCO minival, 40 classes + 40 classes | AP: 17.2 -> 36.9 | Helped |
| use ERD rather than fine-tuning on new data only | GFLV1 · COCO minival, 50 classes + 30 classes | AP: 14.1 -> 36.6 | Helped |
| use ERD rather than fine-tuning on new data only | GFLV1 · COCO minival, 60 classes + 20 classes | AP: 9.8 -> 35.8 | Helped |
| use ERD rather than fine-tuning on new data only | GFLV1 · COCO minival, 70 classes + 10 classes | AP: 4.3 -> 34.9 | Helped |
| distill all classification and regression responses without selection (ablation) | incremental GFLV1 · COCO minival, first 40 classes + last 40 classes | AP: 17.8 -> 31.5 | Hurt |
| distill all classification responses only (ablation) | incremental GFLV1 · COCO minival, first 40 classes + last 40 classes | AP: 17.8 -> 23.8 | Helped |
| distill all regression responses only (ablation) | incremental GFLV1 · COCO minival, first 40 classes + last 40 classes | AP: 17.8 -> 13.0 | Hurt |
| select classification responses with ERS rather than use all classification responses (ablation) | incremental GFLV1 · COCO minival, first 40 classes + last 40 classes | AP: 23.8 -> 33.2 | Helped |
| add regression responses selected by ERS to classification ERS (ablation) | incremental GFLV1 · COCO minival, first 40 classes + last 40 classes | AP: 33.2 -> 36.9 | Helped |
| vary ERS thresholds α₁ and α₂ across the tested settings (ablation) | ERD · COCO minival, first 40 classes + last 40 classes | AP: 36.5; 36.8; 36.7; 36.9 | No clear effect |
Closest papers by meaning
- Incremental Object Detection via Future-Aware Decoupled Cross-Head Distillation (CVPR 2026 · similarity 0.89)
- YOLO-IOD: Towards Real Time Incremental Object Detection (AAAI 2026 · similarity 0.89)
- Demystifying Catastrophic Forgetting in Two-Stage Incremental Object Detector (ICML 2025 · similarity 0.88)
- DCA: Dividing and Conquering Amnesia in Incremental Object Detection (AAAI 2025 · similarity 0.87)
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.86)
- Learning Task-Aware Language-Image Representation for Class-Incremental Object Detection (AAAI 2024 · 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.