Research map / Continual learning / Class-incremental
Preventing Catastrophic Forgetting through Memory Networks in Continuous Detection
Gaurav Bhatt, Leonid Sigal, James Ross
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 memory units to fine-tuned Deformable-DETR (ablation) | MD-DETR · MS-COCO 4-task multi-step | Task 4 mAP@A: 23.7 -> 30.3 | Improved |
| add background thresholding pseudo-labels to FT+Mem (ablation) | MD-DETR · MS-COCO 4-task multi-step | Task 4 mAP@A: 30.3 -> 38.8 | Improved |
| add localized query retrieval to FT+Mem (ablation) | MD-DETR · MS-COCO 4-task multi-step | Task 4 mAP@A: 30.3 -> 42.3 | Improved |
| combine background thresholding with localized query retrieval (ablation) | FT+Mem · MS-COCO 4-task multi-step | Task 4 mAP@A: 42.3 -> 50.2 | Improved |
| lower background threshold to 0.25 (ablation) | MD-DETR · MS-COCO 4-task multi-step | Task 4 mAP@A: 50.2 -> 41.5 | Hurt |
| raise background threshold to 0.85 (ablation) | MD-DETR · MS-COCO 4-task multi-step | Task 4 mAP@A: 50.2 -> 48.3 | Hurt |
| reduce memory capacity or memory length (ablation) | MD-DETR · MS-COCO 4-task multi-step | mAP@P | Hurt |
| enlarge memory to N_m=200 with L_m=20 (ablation) | MD-DETR · MS-COCO 4-task multi-step | mAP@P / mAP@A | Hurt |
| pre-train on COCO+LVIS instead of Task 1 only (ablation) | MD-DETR · MS-COCO 4-task multi-step | mAP@C Task 1 / Task 4 mAP@A: Task 1 78.1 vs 76.2; Task 4 mAP@A 50.2 vs 50.0 | No clear effect |
| replace replay-buffer methods with replay-free memory network | continual detection · PASCAL-VOC 19+1 | mAP@C: 48.5 -> 67.2 | Improved |
Closest papers by meaning
- A continual learning survey: Defying forgetting in classification tasks (TPAMI 2021 · similarity 0.88)
- Alleviating Catastrophic Forgetting of Incremental Object Detection via Within-Class and Between-Class Knowledge Distillation (ICCV 2023 · similarity 0.88)
- Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · similarity 0.87)
- DCA: Dividing and Conquering Amnesia in Incremental Object Detection (AAAI 2025 · similarity 0.87)
- Demystifying Catastrophic Forgetting in Two-Stage Incremental Object Detector (ICML 2025 · similarity 0.87)
- Task-Free Dynamic Sparse Vision Transformer for Continual Learning (AAAI 2024 · similarity 0.86)
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.
