Research map / Detection and segmentation / Weakly supervised detection and segmentation
Rethinking Pseudo Labels for Semi-supervised Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use certainty-aware pseudo labels that estimate localization quality as well as classification confidence (ablation) | semi-supervised object detection · COCO 2% | AP75: 19.4 -> 22.9 | Helped |
| Add class-specific loss reweighting (ablation) | semi-supervised object detection · COCO 2% | AP: 19.9 -> 20.7 | Helped |
| Add dynamic pseudo-label thresholds (ablation) | semi-supervised object detection · COCO 2% | AP: 19.9 -> 20.4 | Helped |
| Combine certainty-aware pseudo labels, loss reweighting, and dynamic thresholding (ablation) | semi-supervised object detection · COCO 2% | AP: 19.9 -> 22.5 | Helped |
| Add color jitter to student training (ablation) | semi-supervised object detection · COCO 2%, student augmentation | AP: 20.3 -> 21.1 | Helped |
| Add Gaussian blur to color jitter (ablation) | semi-supervised object detection · COCO 2%, student augmentation | AP: 21.1 -> 21.5 | Helped |
| Add Cutout to color jitter and Gaussian blur (ablation) | semi-supervised object detection · COCO 2%, student augmentation | AP: 21.5 -> 22.5 | Helped |
| Use only four localization intervals (ablation) | certainty-aware localization · COCO 2% | AP: 22.5 (K=30) -> 19.7 (K=4) | Hurt |
| Add class-specific thresholding and loss reweighting | semi-supervised object detection · COCO 2%, rarest 10 classes | AP: 23.9 -> 26.0 | Helped |
| Use an ensemble of models to generate and evaluate predictions | Ours · COCO 2% | AP: 22.5 -> 23.0 | Helped |
| Sample labeled and unlabeled images at a 1:1 ratio | Ours · PASCAL VOC + COCO-20 | AP: 56.1 -> 56.9 | Helped |
| Apply the proposed method instead of the supervised baseline | Faster R-CNN with FPN, ResNet-50 · COCO-partial, 10% labeled | AP: 23.86 ± 0.81 -> 32.23 ± 0.14 | Helped |
Closest papers by meaning
- Adapting Object Size Variance and Class Imbalance for Semi-supervised Object Detection (AAAI 2023 · similarity 0.96)
- Consistent-Teacher: Towards Reducing Inconsistent Pseudo-Targets in Semi-Supervised Object Detection (CVPR 2023 · similarity 0.92)
- Mind the Gap: Polishing Pseudo Labels for Accurate Semi-supervised Object Detection (AAAI 2023 · similarity 0.92)
- Ambiguity-Resistant Semi-Supervised Learning for Dense Object Detection (CVPR 2023 · similarity 0.92)
- Multi-clue Consistency Learning to Bridge Gaps Between General and Oriented Object in Semi-supervised Detection (AAAI 2025 · similarity 0.90)
- Sparse Semi-DETR: Sparse Learnable Queries for Semi-Supervised Object Detection (CVPR 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.