Research map / Continual learning / Class-incremental
Label-Efficient Online Continual Object Detection in Streaming Video
Jay Zhangjie Wu, David Junhao Zhang, Wynne Hsu, Mengmi Zhang, Mike Zheng Shou
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 Efficient-CLS (EMA slow learner + pseudo-labeling) | iCaRL(our impl.) · EgoObjects (100% annotation) | FAP: 60.80 -> 67.05 | Improved |
| add Efficient-CLS | iCaRL(our impl.) · OAK (100% annotation) | FAP: 36.14 -> 40.24 | Improved |
| add Efficient-CLS at 25% annotation | DER++ · OAK (25% annotation) | FAP: 37.79 -> 37.93 | Improved |
| add EMA slow learner (no pseudo-labeling) (ablation) | iCaRL · OAK 6.25% annotation | FAP: 23.04 -> 27.84 | Improved |
| add pseudo-labeling by the fast learner itself (no EMA) (ablation) | iCaRL · OAK 12.5% annotation | FAP: 28.76 -> 31.60 | Improved |
| add pseudo-labeling by the fast learner itself (no EMA) (ablation) | iCaRL · OAK 12.5% annotation | F (forgetting): -5.48 -> -4.83 | No clear effect |
| add slow-learner pseudo-labeling on top of EMA (ablation) | iCaRL · OAK 25% annotation | FAP: 34.79 -> 38.36 | Improved |
| add data augmentation on pseudo-labeled frames (ablation) | Efficient-CLS · OAK 12.5% annotation | FAP: 30.32 -> 33.92 | Improved |
| use class-balanced replay buffer instead of random store (ablation) | Efficient-CLS · OAK (100% annotation) | FAP: 26.23 -> 40.24 | Improved |
| add RPN loss in pseudo training (ablation) | Efficient-CLS · OAK 12.5% annotation | FAP: 33.92 -> 33.64 | No clear effect |
| raise pseudo-label confidence threshold from 0.1 to 0.7 (ablation) | Efficient-CLS · OAK 12.5% annotation | FAP: 30.33 -> 33.92 | Improved |
| raise pseudo loss weight from 1.0 to 4.0 (ablation) | Efficient-CLS · OAK 12.5% annotation | FAP: 33.92 -> 30.08 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GDumb — GDumb: A Simple Approach that Questions Our Progress in Continual Learning (ECCV 2020) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PseDet: Revisiting the Power of Pseudo Label in Incremental Object Detection (ICLR 2025 · similarity 0.87)
- LEOD: Label-Efficient Object Detection for Event Cameras (CVPR 2024 · similarity 0.86)
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation (ICLR 2024 · similarity 0.85)
- Revisiting Generative Replay for Class Incremental Object Detection (CVPR 2025 · similarity 0.84)
- Summarizing Stream Data for Memory-Constrained Online Continual Learning (AAAI 2024 · similarity 0.84)
- Cut out and Replay: A Simple yet Versatile Strategy for Multi-Label Online Continual Learning (ICML 2025 · similarity 0.84)
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.
