Research map / Continual learning / Class-incremental
Symbiosis-Inspired Knowledge Distillation for Incremental Object Detection
Mingyue Zeng, De Cheng, Zhipeng Xu, Huaijie Wang, Nannan Wang, Xinbo Gao
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 SpSD and SeSD distillation to baseline pseudo-labeling (ablation) | Deformable DETR IOD baseline · COCO 2017 70+10 | AP: 41.2 -> 44.3 | Improved |
| feed high-IoU symbiotic queries directly into distillation (Raw KD) (ablation) | baseline · COCO 2017 70+10 | AP: 41.2 -> 41.1 | Hurt |
| add SpSD alone (ablation) | baseline · COCO 2017 70+10 | AP: 41.2 -> 42.3 | Improved |
| add SeSD alone (ablation) | baseline · COCO 2017 70+10 | AP: 41.2 -> 43.9 | Improved |
| add SIKD to reduce forgetting (FPP) (ablation) | baseline · COCO 2017 70+10 | FPP AP (lower better): 4.9 -> 1.3 | Improved |
| add Consistent Feature Enhancement (CFE) to SpSD (ablation) | SpSD · COCO 2017 70+10 | AP gain over baseline: +0.8 -> +1.1 | Improved |
| add CFE to SeSD (ablation) | SeSD · COCO 2017 70+10 | AP gain over baseline: +2.5 -> +2.7 | Improved |
| anchor-only distillation without symbiotic regions (ablation) | SpSD · COCO 2017 70+10 | AP gain over baseline: +0.3 | Improved |
| increase beta from 4 to 8 (ablation) | SIKD · COCO 2017 70+10 | Old AP / New AP: Old 44.7 -> 45.2, New 39.1 -> 38.5 | Mixed |
| increase alpha from 0.1 to 10 (ablation) | SIKD · COCO 2017 70+10 | Old AP / New AP / Avg: Old 42.1 -> 43.2, New 39.9 -> 34.9, Avg 41.0 -> 39.1 | Mixed |
| use SIKD instead of replay-based CL-DETR* | DIOR · DIOR 19+1 | AP50 All: 57.1 -> 66.1 | Improved |
Closest papers by meaning
- GCD: Advancing Vision-Language Models for Incremental Object Detection via Global Alignment and Correspondence Distillation (AAAI 2025 · similarity 0.90)
- Objects in Semantic Topology (ICLR 2022 · similarity 0.89)
- Incremental Object Detection via Future-Aware Decoupled Cross-Head Distillation (CVPR 2026 · similarity 0.89)
- Alleviating Catastrophic Forgetting of Incremental Object Detection via Within-Class and Between-Class Knowledge Distillation (ICCV 2023 · similarity 0.89)
- YOLO-IOD: Towards Real Time Incremental Object Detection (AAAI 2026 · similarity 0.89)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · 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.
