Research map / Vision-language and multimodal / Cross-modal learning
Deep Global-sense Hard-negative Discriminative Generation Hashing for Cross-modal Retrieval
Kun Cheng, Qibing Qin, Wenfeng Zhang, Lei Huang, Jie Nie
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 the three generation losses (interpolation similarity, semantic preservation, parameter diversification) together (ablation) | DGHDGH · MIRFLICKR-25K, avg over 16-128 bits I2T | mAP (%): 81.64 -> 86.43 | Improved |
| add the three generation losses together (ablation) | DGHDGH · MIRFLICKR-25K, avg over 16-128 bits T2I | mAP (%): 80.04 -> 84.52 | Improved |
| use all three losses instead of the best two-loss combination (84.44 at 16 bits) (ablation) | DGHDGH · MIRFLICKR-25K I2T 16 bits | mAP (%): 84.44 -> 84.66 | Improved |
| use DGHDGH instead of DDBH, I2T, 128 bits | DGHDGH · NUS-WIDE I2T 128 bits | mAP@all (%): 72.29 -> 73.76 | Improved |
| use DGHDGH instead of DDBH, I2T, 128 bits | DGHDGH · MS COCO I2T 128 bits | mAP@all (%): 78.24 -> 79.19 | Improved |
| use DGHDGH instead of DDBH, T2I, 16 bits | DGHDGH · MS COCO T2I 16 bits | mAP@all (%): 71.67 -> 71.16 | Hurt |
| use DGHDGH instead of DDBH under noise 0.8, 16 bits | DGHDGH · MIRFLICKR-25K I2T 16 bits noise 0.2 | mAP (%): 83.79 -> 81.95 | Hurt |
| use DGHDGH instead of DDBH under label noise 0.8, 128 bits | DGHDGH · MIRFLICKR-25K I2T 128 bits noise 0.8 | mAP (%): 80.29 -> 84.16 | Improved |
| swap backbone from CLIP ViT-B/32 to CLIP ViT-L/14 | DGHDGH · MIRFLICKR-25K I2T 16 bits | mAP (%): 84.66 -> 84.82 | No clear effect |
| swap backbone from CLIP ViT-B/32 to SigLIP2 Large-16/256 | DGHDGH · MIRFLICKR-25K I2T 16 bits | mAP (%): 84.66 -> 84.50 | No clear effect |
| swap backbone from CLIP ViT-B/32 to BLIP I-C-Base | DGHDGH · MIRFLICKR-25K I2T 16 bits | mAP (%): 84.66 -> 82.96 | Hurt |
Closest papers by meaning
- Pairwise Relationship Guided Deep Hashing for Cross-Modal Retrieval (AAAI 2017 · similarity 0.91)
- Unsupervised Contrastive Cross-modal Hashing (TPAMI 2022 · similarity 0.91)
- Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval (ICCV 2019 · similarity 0.90)
- Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · similarity 0.89)
- Dual Self-Paced Cross-Modal Hashing (AAAI 2024 · similarity 0.89)
- An End-To-End Graph Attention Network Hashing for Cross-Modal Retrieval (NeurIPS 2024 · similarity 0.88)
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.
