Research map / Continual learning / Class-incremental
Bridge Past and Future: Overcoming Information Asymmetry in Incremental Object Detection
Qijie Mo, Yipeng Gao, Shenghao Fu, Junkai Yan, Ancong Wu, Wei-Shi Zheng
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 Bridge the Past (old-model pseudo labels) (ablation) | BPF baseline with unbiased distillation · VOC 2007 10-10 | mAP@0.5 (1-20): 65.3 -> 71.7 | Improved |
| add Bridge the Future (exclude salient regions from negatives) (ablation) | BPF baseline with unbiased distillation · VOC 2007 10-10 | mAP@0.5 (1-20): 65.3 -> 67.1 | Improved |
| combine Bridge Past and Bridge Future (ablation) | BPF · VOC 2007 10-10 | mAP@0.5 (1-20): 71.7 -> 72.3 | Improved |
| add Distillation with Future (intermediate expert teacher) (ablation) | BPF · VOC 2007 10-10 | mAP@0.5 (1-20): 72.3 -> 72.9 | Improved |
| use expert model as primary teacher on regions overlapping current GT (lambda2=0.5 vs 1.0) (ablation) | Distillation with Future · VOC 2007 10-10 | mAP@0.5 (1-20): 72.4 -> 72.9 | Improved |
| distill all boxes instead of only primary-class boxes (ablation) | Distillation with Future · VOC 2007 10-10 | mAP@0.5 (1-20): 72.9 -> 72.9 | No clear effect |
| add Bridge the Future when more future classes exist (ablation) | BPF · VOC 2007 5-15 | mAP@0.5 (1-20): 72.4 -> 73.0 | Improved |
| combine RPN objectness and attention scores to select future objects (ablation) | Bridge the Future · VOC 2007 10-10 | mAP@0.5 (1-20): 72.3 -> 72.9 | Improved |
| add 2,000-instance memory with copy-paste to BPF | BPF · VOC 2007 10-1 | mAP@0.5 on new classes 11-20: 48.3 -> 57.4 | Improved |
| replace ABR (with memory) by BPF without memory in few-class-per-step settings | BPF · VOC 2007 10-1 and 10-2 | mAP@0.5 (1-20): BPF 55.2 vs ABR 58.9 (10-1) | Hurt |
| replace MMA with BPF | incremental detectors · VOC 2007 5-5 | mAP@0.5 (1-20): 38.9 -> 62.5 | Improved |
Closest papers by meaning
- Alleviating Catastrophic Forgetting of Incremental Object Detection via Within-Class and Between-Class Knowledge Distillation (ICCV 2023 · similarity 0.91)
- Incremental Object Detection via Meta-Learning (TPAMI 2021 · similarity 0.90)
- Incremental Object Detection via Future-Aware Decoupled Cross-Head Distillation (CVPR 2026 · similarity 0.90)
- DCA: Dividing and Conquering Amnesia in Incremental Object Detection (AAAI 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)
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.
