Research map / Detection and segmentation / CNN features
Active Learning for Deep Detection Neural Networks
Hamed H. Aghdam, Abel Gonzalez-Garcia, Joost van de Weijer, Antonio M. Lopez
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 |
|---|---|---|---|
| select frames with the proposed local-divergence score instead of random (guided random) | Active learning detector · Caltech Pedestrian | miss rate at FPPI=1 | Improved |
| select frames with the proposed score instead of random at the first cycle | Active learning detector · Caltech Pedestrian (cycle 1) | miss rate | No clear effect |
| replace local-divergence score with binary entropy (ablation) | Proposed pixel score · Caltech Pedestrian | miss rate at FPPI=1 | Hurt |
| use local-divergence score instead of MC-Dropout (ablation) | MC-Dropout score · Caltech Pedestrian | miss rate at FPPI=1 | Improved |
| use local-divergence score instead of binary entropy for pedestrian instance coverage (ablation) | Binary entropy score · Caltech Pedestrian (7K frames) | pedestrian instances selected: 5243 -> 5706 | Mixed |
| increase budget per cycle from 500 to 1500 (ablation) | Active learning detector · Caltech Pedestrian | miss rate at FPPI=1 | Hurt |
| add max-pooling aggregation instead of averaging pixel scores (ablation) | Active learning detector · Caltech Pedestrian | miss rate at FPPI=1 | Improved |
| add temporal selection rules (ablation) | Active learning detector · Caltech Pedestrian | miss rate at FPPI=1 | Improved |
| add temporal smoothing of image scores (ablation) | Active learning detector · Caltech Pedestrian | miss rate at FPPI=1 | No clear effect |
| use active learning instead of random selection on still images | Active learning detector · BDD100K | miss rate at FPPI=1 | Mixed |
| increase budget per cycle from 50 to 500 images | Active learning cycle · Caltech Pedestrian (B=7500) | time-to-completion (hours): 1800 -> 190 | Improved |
Closest papers by meaning
- Not All Labels Are Equal: Rationalizing the Labeling Costs for Training Object Detection (CVPR 2022 · similarity 0.91)
- Active Learning for Convolutional Neural Networks: A Core-Set Approach (ICLR 2018 · similarity 0.88)
- Active Learning for Deep Object Detection via Probabilistic Modeling (ICCV 2021 · similarity 0.88)
- Towards Human-Machine Cooperation: Self-Supervised Sample Mining for Object Detection (CVPR 2018 · similarity 0.86)
- Active Learning Strategies for Weakly-Supervised Object Detection (ECCV 2022 · similarity 0.86)
- ALWOD: Active Learning for Weakly-Supervised Object Detection (ICCV 2023 · similarity 0.86)
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.
