Research map / Detection and segmentation / CNN features
Django: Detecting Trojans in Object Detection Models via Gaussian Focus Calibration
Guangyu Shen, Siyuan Cheng, Guanhong Tao, Kaiyuan Zhang, Yingqi Liu, Shengwei An, Shiqing Ma, Xiangyu Zhang
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 Gaussian Focus Loss (remove it: DJANGO - GF Loss) (ablation) | DJANGO · Traffic Sign Synthesis SSD | ROC-AUC: 0.7500 -> 1.0000 | Improved |
| add Gaussian Focus Loss (ablation) | DJANGO · Traffic Sign Synthesis Faster-RCNN | ROC-AUC: 0.7500 -> 0.8750 | Improved |
| add Gaussian Focus Loss (ablation) | DJANGO · Traffic Sign Synthesis DETR | ROC-AUC: 0.7500 -> 0.8750 | Improved |
| add label proposal pre-processing (ablation) | DJANGO · Traffic Sign Synthesis SSD | ROC-AUC: 0.6250 -> 1.0000 | Improved |
| add label proposal pre-processing (ablation) | DJANGO · Traffic Sign Synthesis Faster-RCNN | ROC-AUC: 0.5000 -> 0.8750 | Improved |
| add label proposal pre-processing (ablation) | DJANGO · Traffic Sign Synthesis SSD | overhead (s): >3600 -> 859.59 | Improved |
| Gaussian Focus Loss on scanning overhead (ablation) | DJANGO · Traffic Sign Synthesis SSD | overhead (s): 860.24 -> 859.59 | No clear effect |
| warm-up pre-processing added to NC, Pixel, Tabor | NC/Pixel/Tabor · Traffic Sign Synthesis SSD | ROC-AUC: 0.6250 -> 0.6641 | Improved |
| warm-up pre-processing added to Pixel | Pixel · COCO SSD | ROC-AUC: 0.6200 -> 0.7000 | Improved |
| warm-up pre-processing added to NC | NC · DOTA F-RCNN | ROC-AUC: 0.5625 -> 0.5625 | No clear effect |
| DJANGO vs trigger inversion baselines (average) | DJANGO · all | ROC-AUC: 0.592 -> 0.913 | Improved |
| DJANGO vs meta classifier MNTD | DJANGO · COCO SSD | ROC-AUC: 0.8144 -> 0.9305 | Improved |
Closest papers by meaning
- Test-Time Backdoor Detection for Object Detection Models (CVPR 2025 · similarity 0.88)
- Lie Detector: Unified Backdoor Detection via Cross-Examination Framework (NeurIPS 2025 · similarity 0.87)
- Scanning Trojaned Models Using Out-of-Distribution Samples (NeurIPS 2024 · similarity 0.85)
- Energy-based Backdoor Defense without Task-Specific Samples and Model Retraining (ICML 2024 · similarity 0.84)
- RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance (NeurIPS 2026 · similarity 0.84)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · 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.
