Research map / Vision-language and multimodal / Cross-modal learning
Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval
Shupeng Su, Zhisheng Zhong, Chao 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 |
|---|---|---|---|
| merge image and text affinity matrices by weighted sum instead of image affinity alone (ablation) | DJSRH-1 · NUS-WIDE I->T (128 bits) | mAP@50: 0.741 -> 0.747 | Improved |
| merge affinities by weighted sum instead of text affinity alone (ablation) | DJSRH-2 · NUS-WIDE T->I (128 bits) | mAP@50: 0.581 -> 0.738 | Improved |
| add high order neighborhood information (eta = 0.4) (ablation) | DJSRH-3 · NUS-WIDE I->T (128 bits) | mAP@50: 0.747 -> 0.803 | Improved |
| add high order neighborhood information (eta = 0.4) (ablation) | DJSRH-3 · NUS-WIDE T->I (128 bits) | mAP@50: 0.738 -> 0.757 | Improved |
| add regulation parameter mu = 1.5 (ablation) | DJSRH-4 · NUS-WIDE I->T (128 bits) | mAP@50: 0.803 -> 0.812 | Improved |
| add regulation parameter mu = 1.5 (ablation) | DJSRH-4 · NUS-WIDE T->I (128 bits) | mAP@50: 0.757 -> 0.768 | Improved |
| add intra-modal reconstruction terms (lambda1 = lambda2 = 0.1) (ablation) | DJSRH-5 · NUS-WIDE T->I (128 bits) | mAP@50: 0.768 -> 0.789 | Improved |
| add intra-modal reconstruction terms (lambda1 = lambda2 = 0.1) (ablation) | DJSRH-5 · NUS-WIDE I->T (128 bits) | mAP@50: 0.812 -> 0.817 | Improved |
| use tightening tanh (increasing alpha) instead of constant tanh (ablation) | DJSRH · NUS-WIDE I->T (128 bits) | mAP@50: 0.811 -> 0.817 | Improved |
| use tightening tanh (increasing alpha) instead of constant tanh (ablation) | DJSRH · NUS-WIDE T->I (128 bits) | mAP@50: 0.782 -> 0.789 | Improved |
| reconstruct specific similarity values instead of Laplacian constraint (DJSRH-1 vs UDCMH) | DJSRH-1 · NUS-WIDE I->T (128 bits) | mAP@50: 0.558 -> 0.741 | Improved |
| use DJSRH instead of UDCMH on small dataset | DJSRH · Wiki T->I (128 bits) | mAP@50: 0.658 -> 0.658 | No clear effect |
Closest papers by meaning
- Pairwise Relationship Guided Deep Hashing for Cross-Modal Retrieval (AAAI 2017 · similarity 0.93)
- Unsupervised Dual Deep Hashing With Semantic-Index and Content-Code for Cross-Modal Retrieval (TPAMI 2024 · similarity 0.93)
- An Efficient Regenerated Cross-Modal Hashing: Improving Existing Hash Codes With the Arbitrary Length (TPAMI 2026 · similarity 0.93)
- MTFH: A Matrix Tri-Factorization Hashing Framework for Efficient Cross-Modal Retrieval (TPAMI 2019 · similarity 0.92)
- Cross-Modal Deep Variational Hashing (ICCV 2017 · similarity 0.92)
- Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · similarity 0.90)
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.
