Research map / Continual learning / Class-incremental
Continual Detection Transformer for Incremental Object Detection
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 |
|---|---|---|---|
| Replace direct KD and ER with CL-DETR | Deformable DETR · COCO 2017 70+10 | AP: 36.9 -> 40.1 | Helped |
| Add classical knowledge distillation (ablation) | Deformable DETR · COCO 2017 70+10 | AP: 4.2 -> 24.5 | Helped |
| Add joint bipartite matching to classical KD (ablation) | Deformable DETR with classical KD · COCO 2017 70+10 | AP: +5.8 percentage points | Helped |
| Add selected old-category pseudo-labels, merging them with new-category ground truth (ablation) | Deformable DETR with joint matching · COCO 2017 70+10 | AP: 30.3 -> 33.9 | Helped |
| Add distribution-preserving exemplar calibration instead of conventional ER (ablation) | CL-DETR · COCO 2017 70+10 | AP, all categories: +1.8 percentage points | Helped |
| Add distribution-preserving exemplar calibration instead of conventional ER (ablation) | CL-DETR · COCO 2017 70+10 | FPP: 2.1 percentage points | Helped |
| Use top-K selection with K=10 rather than K=5 (ablation) | CL-DETR · COCO 2017 70+10 | AP: 39.7 -> 40.1 | Helped |
| Use top-K selection with K=10 rather than K=20 (ablation) | CL-DETR · COCO 2017 70+10 | AP: 39.9 -> 40.1 | Helped |
| Replace top-K selection with fixed confidence thresholds (ablation) | CL-DETR · COCO 2017 70+10 | AP: 40.1 vs 39.3; 39.6; 39.2 | Hurt |
| Decrease the pseudo-label score threshold from 0.5 to 0.1 during training (ablation) | CL-DETR · COCO 2017 70+10 | AP: +0.3 AP | Helped |
| Add exemplar replay (ablation) | CL-DETR · COCO 2017 40+40, traditional protocol | AP: 39.2 -> 42.0 | Helped |
| Add exemplar replay (ablation) | CL-DETR · COCO 2017 70+10, traditional protocol | AP: 35.8 -> 40.4 | Helped |
Closest papers by meaning
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.89)
- DCA: Dividing and Conquering Amnesia in Incremental Object Detection (AAAI 2025 · similarity 0.89)
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.88)
- Demystifying Catastrophic Forgetting in Two-Stage Incremental Object Detector (ICML 2025 · similarity 0.87)
- GCD: Advancing Vision-Language Models for Incremental Object Detection via Global Alignment and Correspondence Distillation (AAAI 2025 · similarity 0.87)
- SDDGR: Stable Diffusion-based Deep Generative Replay for Class Incremental Object Detection (CVPR 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.