Research map / Detection and segmentation / Open-set and few-shot detection
Enhancing Source-Free Domain Adaptive Object Detection with Low-confidence Pseudo Label Distillation
Ilhoon Yoon, Hyeongjun Kwon, Jin Kim, Junyoung Park, Hyunsung Jang, Kwanghoon Sohn
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 low-confidence pseudo labels (LPL) to high-confidence pseudo labels (HPL) (ablation) | Mean-Teacher SFOD · Cityscapes -> Foggy Cityscapes | mAP: 34.3 -> 38.9 | Improved |
| add feature-distance adaptive weights to LPLD loss (ablation) | HPL+LPL · Cityscapes -> Foggy Cityscapes | mAP: 38.9 -> 40.4 | Improved |
| use KL-divergence for classification and drop regression loss in LPLD instead of cross-entropy plus regression (ablation) | LPLD loss · Cityscapes -> Foggy Cityscapes | mAP: 36.6 -> 38.9 (without adaptive weights) | Improved |
| add regression loss on top of cross-entropy for LPL (ablation) | LPLD loss · Cityscapes -> Foggy Cityscapes | mAP: 37.2 -> 36.6 | Hurt |
| lower IoU threshold of LPL mining against HPL from 0.5 to 0.4 (ablation) | LPL mining · Cityscapes -> Foggy Cityscapes | mAP: 38.3 -> 40.4 | Improved |
| apply background confidence threshold 0.99 to LPL mining instead of none (1.00) (ablation) | LPL mining · Cityscapes -> Foggy Cityscapes | mAP: 39.3 -> 40.4 | Improved |
| use class-confidence threshold 0.9 for LPL instead of 0.4 (ablation) | LPL mining · Cityscapes -> Foggy Cityscapes | mAP: 38.1 -> 40.4 | Improved |
| apply LPLD loss to Mean-Teacher | Mean-Teacher SFOD · Cityscapes -> Foggy Cityscapes | false negatives: 11491 -> 2822 false negatives | Improved |
| apply LPLD on Sim10k -> Cityscapes against PETS | SFOD · Sim10k -> Cityscapes | AP50 of car: 57.8 (PETS) vs 49.4 | Hurt |
| apply LPLD on Pascal-VOC -> Clipart | SFOD · Pascal-VOC -> Clipart | mAP: 31.5 -> 34.0 (vs IRG) | Improved |
Closest papers by meaning
- A Free Lunch for Unsupervised Domain Adaptive Object Detection without Source Data (AAAI 2021 · similarity 0.94)
- Simplifying Source-Free Domain Adaptation for Object Detection: Effective Self-Training Strategies and Performance Insights (ECCV 2024 · similarity 0.93)
- Large Self-Supervised Models Bridge the Gap in Domain Adaptive Object Detection (CVPR 2025 · similarity 0.90)
- Non-exemplar Domain Incremental Object Detection via Learning Domain Bias (AAAI 2024 · similarity 0.89)
- Active Domain Adaptation with False Negative Prediction for Object Detection (CVPR 2024 · similarity 0.89)
- CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection (CVPR 2024 · similarity 0.89)
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.
