Research map / Vision-language and multimodal / Cross-modal learning
MTFH: A Matrix Tri-Factorization Hashing Framework for Efficient Cross-Modal Retrieval
Xin Liu, Zhikai Hu, Haibin Ling, Yiu‐ming Cheung
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 unequal hash lengths (text 64 bits, image 128 bits) instead of equal 128 bits (ablation) | MTFH · Wiki I→T recall top-500 | recall: 0.3876 -> 0.4093 | Improved |
| shrink text code to 16 bits with 128-bit image code instead of 128 bits (ablation) | MTFH · Wiki I→T recall top-500 | recall: 0.3876 -> 0.3041 | Hurt |
| use unequal code length 32 for text instead of equal 128 at top-2000 (ablation) | MTFH · Wiki I→T recall top-2000 | recall: 0.9554 -> 0.9640 | Improved |
| k-means instead of random sampling in kernel logistic regression (ablation) | MTFH · NUS-WIDE-All I→T (128 bits) | mAP: 0.6228 -> 0.6271 | Improved |
| k-means instead of random sampling in kernel logistic regression (ablation) | MTFH · Wiki I→T (128 bits) | mAP: 0.3388 -> 0.3349 | Hurt |
| k-means instead of random sampling in kernel logistic regression (ablation) | MTFH · MIRFlickr T→I (128 bits) | mAP: 0.8303 -> 0.8352 | Improved |
| MTFH versus DCH (joint hash function and unified code learning), single-label data | MTFH_km · Wiki I→T (128 bits) | mAP: 0.3783 -> 0.3349 | Hurt |
| MTFH versus DCH on multi-label large dataset | MTFH_km · NUS-WIDE-All T→I (128 bits) | mAP: 0.5298 -> 0.7307 | Improved |
| MTFH versus SePH on multi-label dataset | MTFH_km · MIRFlickr I→T (128 bits) | mAP: 0.6851 -> 0.7676 | Improved |
| MTFH versus DCH at 32 bits, top-50 precision | TMFH_rnd · NUS-WIDE-100k I→T | mAP@50: 0.6396 -> 0.7337 | Improved |
| MTFH versus DCH at 32 bits, top-50 precision | MTFH · MIRFlickr T→I (32 bits) | mAP@50: 0.8923 -> 0.8624 | Hurt |
| use unequal code lengths 80 and 100 instead of 100 and 100 (ablation) | MTFH · MIRFlickr | mAP | Improved |
Closest papers by meaning
- Pairwise Relationship Guided Deep Hashing for Cross-Modal Retrieval (AAAI 2017 · similarity 0.93)
- An Efficient Regenerated Cross-Modal Hashing: Improving Existing Hash Codes With the Arbitrary Length (TPAMI 2026 · similarity 0.93)
- Unsupervised Dual Deep Hashing With Semantic-Index and Content-Code for Cross-Modal Retrieval (TPAMI 2024 · similarity 0.92)
- Dynamic Masking and Auxiliary Hash Learning for Enhanced Cross-Modal Retrieval (NeurIPS 2025 · similarity 0.92)
- Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · similarity 0.92)
- Asymmetric Cross-Modal Hashing Based on Formal Concept Analysis (AAAI 2025 · similarity 0.92)
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.
