Research map / Continual learning / Class-incremental
DuET: Dual Incremental Object Detection via Exemplar-Free Task Arithmetic
Munish Monga, Vishal Chudasama, Pankaj Wasnik, Biplab Banerjee
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 sequential fine-tuning over independent fine-tuning (ablation) | DuET · VOC [1:10] -> Clipart [11:20] | RAI (%): 4.82 -> 6.81 | Improved |
| add Incremental Head (ablation) | DuET with sequential FT · VOC [1:10] -> Clipart [11:20] | RAI (%): 6.81 -> 29.06 | Improved |
| add DuET task-arithmetic module (ablation) | DuET with Incremental Head · VOC [1:10] -> Clipart [11:20] | RAI (%): 29.06 -> 56.13 | Improved |
| add distillation loss L_Distill* (ablation) | DuET · VOC [1:10] -> Clipart [11:20] | RAI (%): 56.13 -> 62.41 | Improved |
| add directional consistency loss L_DC (ablation) | DuET · VOC [1:10] -> Clipart [11:20] | RAI (%): 62.41 -> 65.99 | Improved |
| replace L_Distill with modified L_Distill* (ablation) | DuET · VOC [1:10] -> Clipart [11:20] | RAI (%): 53.19 -> 62.41 | Improved |
| add L_DC to Weight-Averaging merging (ablation) | Weight-Averaging · VOC [1:10] -> Clipart [11:20] | RAI (%): 48.92 -> 56.83 | Improved |
| add L_DC to MagMax merging (ablation) | MagMax · VOC [1:10] -> Clipart [11:20] | RAI (%): 46.57 -> 47.54 | Improved |
| add L_DC to Fisher-Merging (ablation) | Fisher-Merging · VOC [1:10] -> Clipart [11:20] | RAI (%): 18.71 -> 22.82 | Improved |
| add L_DC to EMR-Merging (ablation) | EMR-Merging · VOC [1:10] -> Clipart [11:20] | RAI (%): 51.03 -> 52.25 | Improved |
| add L_DC across task-arithmetic approaches: sign conflicts (ablation) | task arithmetic merging methods · VOC [1:10] -> Clipart [11:20] | sign conflicts: 34% average reduction | Improved |
| use YOLO11n instead of larger detectors as base detector | DuET · Daytime Sunny [1:4] -> Night Sunny [5:7] | RAI (%): 27.89 -> 72.51 | Mixed |
Closest papers by meaning
- Orthogonal Knowledge Refreshing for Domain-Incremental Object Detection (ECCV 2026 · similarity 0.88)
- Boosting Vision-Language Models Towards Cross-Domain Incremental Object Detection (CVPR 2026 · similarity 0.88)
- Real-Time Source-Free Object Detection (ECCV 2026 · similarity 0.88)
- YOLO-IOD: Towards Real Time Incremental Object Detection (AAAI 2026 · similarity 0.88)
- Non-exemplar Domain Incremental Object Detection via Learning Domain Bias (AAAI 2024 · similarity 0.87)
- DCA: Dividing and Conquering Amnesia in Incremental 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.
