Research map / Adversarial robustness and security / Attacks on text and graphs
Using LIP to Gloss Over Faces in Single-Stage Face Detection Networks
Siqi Yang, Arnold Wiliem, Shaokang Chen, Brian C. Lovell
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 |
|---|---|---|---|
| replace whole-image perturbation (IMP) with ERF-cropped localized perturbation (LP) | I-FGSM · FDDB | attack success rate (%): 53.9 -> 94.9 | Improved |
| replace IMP with LIP-H | I-FGSM · FDDB | attack success rate (%): 53.9 -> 93.8 | Improved |
| compute perturbation per instance before cropping (LIP-H) instead of jointly (LP) (ablation) | I-FGSM · FDDB | attack success rate (%): 94.9 -> 93.8 | No clear effect |
| compute perturbation per instance before cropping (LIP-H) instead of jointly (LP) (ablation) | I-FGSM · WIDER FACE val (easy) | attack success rate (%): 67.4 -> 71.3 | Improved |
| compute perturbation per instance before cropping (LIP-H) instead of jointly (LP) (ablation) | I-FGSM · WIDER FACE val (medium) | attack success rate (%): 61.7 -> 65.7 | Improved |
| compute perturbation per instance before cropping (LIP-A) instead of jointly (LP) (ablation) | I-FGSM · WIDER FACE val (hard) | attack success rate (%): 62.1 -> 69.5 | Improved |
| replace IMP with LIP-A | DeepFool · WIDER FACE val (hard) | attack success rate (%): 39.6 -> 66.6 | Improved |
| replace IMP with LIP-A | DeepFool · WIDER FACE val (easy) | attack success rate (%): 45.3 -> 53.2 | Improved |
| attack highest-confidence proposal only (LIP-H) instead of all proposals (LIP-A) (ablation) | LIP · WIDER FACE val (easy) | attack success rate (%): 69.5 -> 71.3 | No clear effect |
| attack highest-confidence proposal only (LIP-H) instead of all proposals (LIP-A) (ablation) | LIP · WIDER FACE val (hard) | attack success rate (%): 69.5 -> 65.6 | Hurt |
| increase number of faces from 1 to 81 at distance 40 | IMP(I-FGSM) · Synthetic faces | attack success rate (%): 100 -> 18.3 | Hurt |
| increase distance between faces from 40 to 160 (16 faces) | IMP(I-FGSM) · Synthetic faces | attack success rate (%): 34 -> 38.9 | Improved |
Closest papers by meaning
- Towards Certifiably Robust Face Recognition (ECCV 2024 · similarity 0.84)
- Design and Interpretation of Universal Adversarial Patches in Face Detection (ECCV 2020 · similarity 0.84)
- LipSim: A Provably Robust Perceptual Similarity Metric (ICLR 2024 · similarity 0.83)
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.83)
- Once and for All: Universal Transferable Adversarial Perturbation against Deep Hashing-Based Facial Image Retrieval (AAAI 2024 · similarity 0.83)
- SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining (AAAI 2025 · similarity 0.82)
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.
