Research map / Adversarial robustness and security / Backdoor and poisoning
Phantom: Physical Object Interactions as Dynamic Triggers for NMS-Exploited Backdoors
Tianlin Huo, Dongchuan Ran, Ranjie Duan, Yao Zhu, Peilun Du, Ningbo Yao, Huanqian Yan, Xu Han, Qiang Yun, Yuzheng Tan, Yang Bao, Yuan He
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 |
|---|---|---|---|
| vary poisonous target box size from 0.7x to 1.3x (ablation) | Phantom · COCO and VOC (YOLOv5 setting) | ASR: ASR above 90% | No clear effect |
| shift poisonous box position by -30% to +30% (ablation) | Phantom · COCO and VOC | ASR: ASR exceeding 95% | No clear effect |
| lower victim/target training confidence combination below (0.7, 0.5) (ablation) | Phantom · COCO and VOC | ASR: ASR zero | Hurt |
| raise victim and target training confidence (ablation) | Phantom · COCO and VOC | victim detection rate | Improved |
| raise IoU threshold gamma above 0.2 (ablation) | Phantom · COCO and VOC | ASR: ASR stable above 90% in range 0.2 to 1.0 | Improved |
| apply Gaussian blur input transformation defense | Phantom · COCO and VOC | ASR | No clear effect |
| scan Phantom models with ODSCAN | Phantom · COCO and VOC | ASR under ODSCAN: ASR below threshold 0.9 | No clear effect |
| scan Phantom models with Detector Cleanse | Phantom · MS-COCO, YOLOv5 | FRR/FAR: FAR 93.53 (MCA) vs 93.35 (clean) at baseline offset | No clear effect |
| use Phantom instead of GMA for misclassification | Faster R-CNN · MS-COCO benign | mAP50: 53.76 -> 59.13 | Improved |
| use Phantom object appearing attack instead of Clean-label on YOLOv5 | YOLOv5 · VOC poisoned | ASR: 94.88 -> 55.00 | Hurt |
| use Phantom instead of ODA/UT/Clean-label for object disappearing | Faster R-CNN · COCO poisoned | ASR: 85.83 -> 91.84 (vs Clean-label) | Improved |
Closest papers by meaning
- Detecting Backdoors in Object Detection via Pre-NMS Prediction Distribution Shift (ECCV 2026 · similarity 0.90)
- Finding Naturally Occurring Physical Backdoors in Image Datasets (NeurIPS 2022 · similarity 0.89)
- Test-Time Backdoor Detection for Object Detection Models (CVPR 2025 · similarity 0.87)
- Backdoor Attacks on Open Vocabulary Object Detectors via Multi-Modal Prompt Tuning (AAAI 2026 · similarity 0.87)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.86)
- A Closer Look at Backdoor Attacks on CLIP (ICML 2025 · similarity 0.85)
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.
