Research map / Continual learning / Class-incremental
Incremental Object Detection via Meta-Learning
K J Joseph, Jathushan Rajasegaran, Salman Hameed Khan, Fahad Shahbaz Khan, Vineeth Nallure Balasubramanian
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 fine-tuning on image store (ablation) | distillation-only detector · VOC07 15+5 | mAP: 58.5 -> 63.6 | Improved |
| add fine-tuning (ablation) | gradient-preconditioned detector · VOC07 15+5 | mAP: 47.6 -> 66.1 | Improved |
| add gradient preconditioning when fine-tuning a distilled model (ablation) | distillation + fine-tuning · VOC07 15+5 | mAP: 63.6 -> 67.8 | Improved |
| add distillation loss (ablation) | preconditioning + fine-tuning · VOC07 15+5 | mAP: 66.1 -> 67.8 | Improved |
| add fine-tuning (ablation) | distillation + preconditioning · VOC07 15+5 | mAP: 54.3 -> 67.8 | Improved |
| raise distillation weight alpha from 0.1 to 0.8 (ablation) | distillation-only detector · VOC07 15+5 | mAP new classes: 40.8 -> 3.0 | Hurt |
| use L2 feature distillation instead of NST (ablation) | distillation · VOC07 15+5 | mAP: 55.5 -> 58.5 | Improved |
| use L2 feature distillation instead of AT (ablation) | distillation · VOC07 15+5 | mAP: 44.2 -> 58.5 | Improved |
| train RPN along with new classes (ablation) | distilled detector · VOC07 15+5 | mAP: 58.1 -> 58.5 | No clear effect |
| put warp layer in last residual block instead of earlier blocks (ablation) | preconditioning · VOC07 15+5 | mAP: 54.1 -> 58.5 | Improved |
| add moving-average (Mean Teacher) teacher (ablation) | Ours · VOC07 10+10 | mAP: 66.3 -> 62.1 | Hurt |
| add moving-average teacher (nu=0.99) (ablation) | Ours · VOC07 15+5 | mAP: 67.8 -> 66.7 | No clear effect |
Closest papers by meaning
- Gradient Decomposition and Alignment for Incremental Object Detection (ICCV 2025 · similarity 0.91)
- Incremental Learning of Object Detectors Without Catastrophic Forgetting (ICCV 2017 · similarity 0.91)
- Focus, Align, and Sustain: Counteracting Gradient Dilution in Incremental Object Detection (ICML 2026 · similarity 0.90)
- Bridge Past and Future: Overcoming Information Asymmetry in Incremental Object Detection (ECCV 2024 · similarity 0.90)
- Incremental Object Detection via Future-Aware Decoupled Cross-Head Distillation (CVPR 2026 · similarity 0.90)
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.88)
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.
