Research map / Vision-language and multimodal / Cross-modal learning
Cross Modal Retrieval With Querybank Normalisation
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 |
|---|---|---|---|
| Apply QB-Norm to the retrieval model | Cross-modal retrieval models · MSR-VTT 1k-A, MSVD, DiDeMo, LSMDC, VaTeX, QuerYD, MSCoCo, CUB-200-2011, Stanford Online Products, AudioCaps | R@1: 29.6 -> 33.3; 25.4 -> 26.6; 21.6 -> 24.2; 17.2 -> 17.8; 53.2 -> 54.8; 14.4 -> 15.1; 30.3 -> 34.8; 64.4 -> 64.8; 77.8 -> | Helped |
| Apply QB-Norm to CLIP2Video | CLIP2Video · MSR-VTT 1k-A and MSVD | R@1: 45.6 -> 47.2; 47.0 -> 47.6 | Helped |
| Use a training-set rather than test-set querybank (ablation) | QB-Norm with DIS · MSR-VTT full | R@1: 17.5 -> 17.3 | No clear effect |
| Use a validation-set rather than test-set querybank (ablation) | QB-Norm with DIS · MSR-VTT full | R@1: 17.5 -> 16.6 | Hurt |
| Use an LSMDC far-domain querybank with GC, CSLS, or IS rather than DIS (ablation) | QB-Norm · MSR-VTT full | R@1: DIS 14.9; GC 14.8; CSLS 13.4; IS 11.6; no querybank 14.9 | Hurt |
| Use DIS rather than IS normalization with an adversarial querybank (ablation) | QB-Norm · MSR-VTT full | R@1: 12.3 -> 14.9 | Helped |
| Increase DIS gallery-activation top-k from 1 to 5 (ablation) | QB-Norm with DIS · MSR-VTT full, in-domain querybank | R@1: 17.0 -> 17.1 | No clear effect |
| Increase DIS gallery-activation top-k from 1 to 5 (ablation) | QB-Norm with DIS · MSR-VTT full, far-domain querybank | R@1: 14.9 -> 14.6 | Hurt |
| Apply QB-Norm to embeddings | Cross-modal retrieval embeddings · MSR-VTT, DiDeMo, LSMDC, MSCoCo | hubness skewness: 0.939 -> 0.509; 1.21 -> 0.39; 0.715 -> 0.321; 0.56 -> 0.16 | Helped |
| Use CENT normalization (ablation) | TT-CE+ · MSR-VTT full | R@1: 15.0 -> 14.4 | Hurt |
Closest papers by meaning
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.87)
- NeighborRetr: Balancing Hub Centrality in Cross-Modal Retrieval (CVPR 2025 · similarity 0.86)
- Learning to Rematch Mismatched Pairs for Robust Cross-Modal Retrieval (CVPR 2024 · similarity 0.86)
- Adaptive Prompt-Based Semantic Embedding with Inspire Potential of Implicit Knowledge for Cross-Modal Retrieval (AAAI 2025 · similarity 0.86)
- Semi-Parametric Retrieval via Binary Bag-of-Tokens Index (ICLR 2025 · similarity 0.85)
- MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS (ICLR 2025 · 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.