Research map / Label-efficient and robust learning / Knowledge distillation
Distilling Object Detectors via Monte Carlo Dropout
Junfei Yi, Hui Zhang, Jianxu Mao, Tengfei Liu, Mingjie Li, Sihao Lin, Hanyu Gu, Zhihui Li, Xiaojun Chang, Yaonan Wang
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 MC-dropout teacher knowledge uncertainty (N=10) | FGD · GFL Res101 to Res50 COCO | mAP: 43.4 -> 44.1 | Improved |
| increase MC dropout samples N from 10 to 15 (ablation) | FGD+Ours · GFL COCO | mAP: 44.1 -> 43.8 | Hurt |
| use a single dropout sample (N=1) (ablation) | FGD+Ours · GFL COCO | mAP: 43.4 -> 43.9 | Improved |
| vary dropout ratio strategy a, b, c (ablation) | FGD+Ours · GFL COCO | mAP: 44.0 -> 44.0 | No clear effect |
| add uncertainty to teacher knowledge only (ablation) | FGD · GFL COCO | mAP: 43.4 -> 44.0 | Improved |
| add uncertainty to student knowledge only (ablation) | FGD · GFL COCO | mAP: 43.4 -> 43.6 | Improved |
| add uncertainty to both teacher and student (ablation) | FGD+Ours · GFL COCO | mAP: 44.0 -> 44.0 | No clear effect |
| add residual structure to uncertainty injection (ablation) | FGD+Ours teacher side · GFL COCO | mAP: 43.5 -> 44.0 | Improved |
| add uncertainty to feature KD MGD | MGD · GFL 1x COCO | mAP: 42.1 -> 43.0 | Improved |
| add uncertainty to feature KD PKD | PKD · GFL 1x COCO | mAP: 42.5 -> 42.6 | Improved |
| add uncertainty to logit KD LD | LD · GFL 1x COCO | mAP: 41.0 -> 41.2 | Improved |
| add uncertainty to FGD on Res18 student | FGD · GFL Res18 student COCO | mAP: 33.3 -> 37.9 | Improved |
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.
- CrossKD — CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024) · 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
- When Object Detection Meets Knowledge Distillation: A Survey (TPAMI 2023 · similarity 0.92)
- Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors (ICLR 2021 · similarity 0.92)
- PKD: General Distillation Framework for Object Detectors via Pearson Correlation Coefficient (NeurIPS 2022 · similarity 0.91)
- Prediction-Guided Distillation for Dense Object Detection (ECCV 2022 · similarity 0.91)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.91)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.90)
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.
