Research map / Continual learning / Class-incremental
Towards Non Co-occurrence Incremental Object Detection with Unlabeled In-the-Wild Data
Na Dong, Yongqiang Zhang, Mingli Ding, Gim Hee Lee
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 R-CNN distillation loss (ablation) | dual-teacher student · VOC 19+1 | mAP all (%): 62.9 -> 66.9 | Improved |
| add image-level RoI mask distillation loss (ablation) | dual-teacher student · VOC 19+1 | mAP all (%): 62.9 -> 67.1 | Improved |
| add instance-level heatmap distillation loss (ablation) | dual-teacher student · VOC 19+1 | mAP all (%): 62.9 -> 66.2 | Improved |
| add R-CNN distillation loss (ablation) | dual-teacher student · VOC 19+1 | mAP novel (%): 35.3 -> 44.9 | Improved |
| add blind sampling strategy to the full distillation framework (ablation) | dual-teacher student · VOC 19+1 | mAP all (%): 68.7 -> 70.2 | Improved |
| add all three distillation losses to pseudo-label training (ablation) | dual-teacher student · VOC 19+1 | mAP all (%): 62.9 -> 70.2 | Improved |
| use in-the-wild data with category overlap instead of without | Ours · VOC 19+1 | mAP all (%): 70.2 -> 72.2 | Improved |
| use in-the-wild data with category overlap instead of without | Ours · VOC 10+10 | mAP all (%): 60.4 -> 59.9 | Hurt |
| use our method instead of Shmelkov without co-occurrence | incremental detector · VOC 19+1 | mAP all (%): 61.4 -> 72.2 | Improved |
| use our method instead of Shmelkov without co-occurrence | incremental detector · VOC 10+10 | mAP novel (%): 49.1 -> 62.2 | Improved |
| use our method instead of Shmelkov without co-occurrence | incremental detector · VOC 10+5+5 base classes | mAP base (%): 59.0 -> 56.7 | Hurt |
| use Open Images as in-the-wild data | Ours · VOC 19+1 | mAP novel (%): 39.2 -> 48.6 | Improved |
Closest papers by meaning
- Bridging Non Co-occurrence with Unlabeled In-the-wild Data for Incremental Object Detection (NeurIPS 2021 · similarity 0.99)
- Incremental Learning of Object Detectors Without Catastrophic Forgetting (ICCV 2017 · similarity 0.91)
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.89)
- Demystifying Catastrophic Forgetting in Two-Stage Incremental Object Detector (ICML 2025 · similarity 0.88)
- DCA: Dividing and Conquering Amnesia in Incremental Object Detection (AAAI 2025 · similarity 0.88)
- Non-exemplar Domain Incremental Object Detection via Learning Domain Bias (AAAI 2024 · 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.
