Research map / Adversarial robustness and security / Certified and empirical robustness
DRFGD: Disentangled Representation-Focused Generative Defense for Attack-Tolerant Cross-Modal Hashing
Zhongqing Yu, Xin Liu, Yiu-ming Cheung, Zhikai Hu, Wentao Fan, Pan Zhou
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 divergence loss LDiv (ablation) | DRFGD · FLICKR-25K I->T (CPAH, TA-DCH) | MAP: 74.26 -> 74.85 | Improved |
| add classification loss LCls (ablation) | DRFGD · FLICKR-25K I->T (CPAH, TA-DCH) | MAP: 73.88 -> 74.85 | Improved |
| add alignment loss LAli (ablation) | DRFGD · MS-COCO T->I (CPAH, TA-DCH) | MAP: 49.79 -> 61.82 | Improved |
| add alignment loss LAli (ablation) | DRFGD · FLICKR-25K T->I (CPAH, TA-DCH) | MAP: 67.83 -> 74.70 | Improved |
| add reconstruction loss LRec (ablation) | DRFGD · MS-COCO T->I (CPAH, TA-DCH) | MAP: 43.95 -> 61.82 | Improved |
| add reconstruction loss LRec (ablation) | DRFGD · FLICKR-25K T->I (CPAH, TA-DCH) | MAP: 61.19 -> 74.70 | Improved |
| add discrimination loss LDis (ablation) | DRFGD · NUS-WIDE T->I (CPAH, TA-DCH) | MAP: 64.90 -> 66.58 | Improved |
| use DRFGD instead of best baseline defense | DADH, EQB2A · FLICKR-25K I->T | MAP: 70.13 -> 72.06 | Improved |
| use DRFGD under targeted TA-DCH attack | DADH · I->T retrieval | MAP: 1.81%, 4.10%, 1.60% | Improved |
| use DRFGD instead of baselines for t-MAP | DAPH · I->T | t-MAP: 2.23%, 1.67%, 2.40% reduction | Improved |
| use DRFGD instead of other defenses on benign data | DAPH · FLICKR-25K I->T benign | MAP: 83.42 -> 84.61 | Improved |
| use adversarial training defenses (vs DRFGD) on benign queries | ATRDH/SAAT · FLICKR-25K I->T benign (CPAH) | MAP: 75.76 -> 82.53 | Hurt |
Closest papers by meaning
- Deep Supervised Adversarial Robust Hashing for Retrieval (TPAMI 2026 · similarity 0.89)
- Unsupervised Generative Adversarial Cross-Modal Hashing (AAAI 2018 · similarity 0.88)
- Deep Global-sense Hard-negative Discriminative Generation Hashing for Cross-modal Retrieval (ICLR 2026 · similarity 0.88)
- Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · similarity 0.85)
- Dual Self-Paced Cross-Modal Hashing (AAAI 2024 · similarity 0.85)
- Once and for All: Universal Transferable Adversarial Perturbation against Deep Hashing-Based Facial Image Retrieval (AAAI 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.
