Research map / Detection and segmentation / CNN features
IoUCert: Robustness Verification for Anchor-based Object Detectors
Benedikt Brückner, Alejandro J. Mercado, Yanghao Zhang, Panagiotis Kouvaros, Alessio Lomuscio
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 |
|---|---|---|---|
| use AvgPool instead of MaxPool downsampling (ablation) | IoUCert on YOLOv3-tiny (LARD 64x64) · LARD brightness eps=0.3 | robust cases (of 50): 15 -> 50 | Improved |
| use AvgPool instead of MaxPool downsampling (ablation) | IoUCert on YOLOv3-tiny (LARD 64x64) · LARD brightness eps=0.3 | mean verification time (s): 1602.32 -> 56.26 | Improved |
| use AvgPool instead of MaxPool downsampling (ablation) | IoUCert on YOLOv3-tiny (LARD 64x64) · LARD contrast eps=0.3 | robust cases (of 50): 6 -> 50 | Improved |
| use AvgPool instead of MaxPool downsampling (ablation) | YOLOv3-tiny (LARD 64x64) · clean detection | mAP@0.5 (%): 86.88 -> 86.59 | No clear effect |
| use AvgPool instead of MaxPool downsampling (ablation) | YOLOv3-tiny (LARD 64x64) · clean detection | mAP@0.5:0.95 (%): 41.67 -> 40.90 | Hurt |
| use tighter optimal IoU bounds instead of Cohen et al. bounds (ablation) | IoUCert on SSD · SSD brightness eps=0.02 bounds | tightness improvement (%): 50.67 -> 65.09 | Improved |
| use tighter optimal IoU bounds instead of Cohen et al. bounds (ablation) | IoUCert on SSD · SSD brightness eps=0.05 | mean verification time (s): 389.59 -> 404.54 | No clear effect |
| use tighter optimal IoU bounds instead of Cohen et al. bounds (ablation) | IoUCert on SSD · SSD brightness eps=0.01 | branches explored: 74 -> 66 | Improved |
| use tighter optimal IoU bounds on shallow IoU ranges (ablation) | IoUCert on SSD · SSD IoU bounds 0.50-0.60 | avoided exploration (%): 99.66 | Improved |
| increase perturbation budget eps from 0.01 to 0.30 | IoUCert on SSD · LARD brightness | robust cases (of 50): 48 -> 9 | Hurt |
| increase input resolution from 64x64 to 128x128 | IoUCert on YOLOv3-tiny · LARD brightness eps=0.5 | robust cases (of 50): 47 -> 40 | Hurt |
| train on COCO instead of LARD at 128x128 | YOLOv3-tiny · brightness eps=0.5 | robust cases (of 50): 40 -> 29 | Hurt |
Closest papers by meaning
- Probabilistic Anchor Assignment with IoU Prediction for Object Detection (ECCV 2020 · similarity 0.86)
- Anchor-Intermediate Detector: Decoupling and Coupling Bounding Boxes for Accurate Object Detection (ICCV 2023 · similarity 0.86)
- CIA-SSD: Confident IoU-Aware Single-Stage Object Detector From Point Cloud (AAAI 2021 · similarity 0.86)
- Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (AAAI 2021 · similarity 0.86)
- Acquisition of Localization Confidence for Accurate Object Detection (ECCV 2018 · similarity 0.85)
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · 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.
