Research map / Vision-language and multimodal / Cross-modal learning
Creating Something From Nothing: Unsupervised Knowledge Distillation for Cross-Modal Hashing
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 teacher-derived pair similarities to guide a supervised student | UKD-SS · MIRFlickr-25K, image→text, 16 bits | mAP: 0.676 -> 0.714 | Helped |
| use teacher-derived pair similarities to guide a supervised student | UKD-SS · NUS-WIDE, text→image, 128 bits | mAP: 0.637 -> 0.663 | Helped |
| use teacher-derived pair similarities to guide an unsupervised student | UKD-US · MIRFlickr, image→text, 16 bits | mAP: 0.676 -> 0.695 | Helped |
| measure pair similarity with teacher image embeddings rather than raw image features (ablation) | UKD similarity estimation · MIRFlickr, P@5000 | precision: 64.4% -> 74.6% | Helped |
| measure pair similarity with teacher text embeddings rather than raw text features (ablation) | UKD similarity estimation · MIRFlickr, P@5000 | precision: 55.6% -> 67.9% | Helped |
| combine image and text similarity directly instead of using image similarity alone (ablation) | UKD similarity estimation · MIRFlickr, P@5000 | precision: 74.6% -> 73.4% | Hurt |
| merge separate image-based and text-based lists of relevant pairs (ablation) | UKD similarity estimation · MIRFlickr, P@5000 | precision: 76.1% | Helped |
| distill a second time from the first student (ablation) | UKD-US · MIRFlickr, image→text | mAP: 0.33% compared to 0.60% of the first round | Helped |
| use a 16-bit teacher instead of the 128-bit teacher (ablation) | UKD-SS · MIRFlickr-25K | mAP | Mixed |
| increase the number of selected relevant pairs (ablation) | UKD · MIRFlickr | mAP | Mixed |
Closest papers by meaning
- Dual Self-Paced Cross-Modal Hashing (AAAI 2024 · similarity 0.91)
- Lightweight Contrastive Distilled Hashing for Online Cross-modal Retrieval (AAAI 2025 · similarity 0.91)
- Semi-Supervised Online Cross-Modal Hashing (AAAI 2025 · similarity 0.90)
- Vision-guided Text Mining for Unsupervised Cross-modal Hashing with Community Similarity Quantization (AAAI 2025 · similarity 0.90)
- SHE: Streaming-media Hashing Retrieval (ICML 2025 · similarity 0.89)
- Robust Self-Paced Hashing for Cross-Modal Retrieval with Noisy Labels (AAAI 2025 · similarity 0.89)
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.