Research map / Vision-language and multimodal / Cross-modal learning
Multi-Lingual Acquisition on Multimodal Pre-training for Cross-modal Retrieval
Liang Zhang, Anwen Hu, Qin Jin
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 |
|---|---|---|---|
| remove Language Exposure stage (ablation) | MLA-CLIP · MSRVTT zero-shot mean | text-to-video R@1: 16.7 -> 17.8 | Improved |
| add LE stage after NLT (stage two) (ablation) | MLA-CLIP · Multi30K de | average recall: 76.3 -> 78.7 | Improved |
| do LE at stage one instead of NLT (ablation) | MLA-CLIP · Multi30K de | average recall: 76.3 -> 68.2 | Hurt |
| joint NLT+LE at stage one (ablation) | MLA-CLIP · Multi30K de | average recall: 76.3 -> 71.1 | Hurt |
| optimize NLT and LE together at stage two (ablation) | MLA-CLIP · Multi30K de | average recall: 78.7 vs 78.4 | No clear effect |
| remove language acquirers (ablation) | MLA-CLIP · Multi30K cs | average recall: 65.7 -> 70.8 | Improved |
| remove M-BERT embedding initialization (ablation) | MLA-CLIP · Multi30K cs | average recall: 69.4 -> 70.8 | Improved |
| compare with translate-then-retrieve (CLIP+MT) | MLA-CLIP · Multi30K de | average recall: 75.3 -> 78.7 | Improved |
| use CLIP ViT-B/16 instead of ViT-B/32 backbone | MLA-CLIP · Multi30K de zero-shot | average recall: 78.7 -> 80.8 | Improved |
| fine-tune only language acquirers on downstream data (FT-All) after FT-En | MLA-CLIP · Multi30K de | average recall: 82.6 -> 86.8 | Improved |
| add text-text plus image-text data for Korean low-resource (600 samples) | MLA-CLIP · XTD Korean | average recall: 78.7 -> 80.1 | Improved |
| train on unseen languages after seen ones | UC2 · XTD en | average recall: 71.8 -> 63.6 | Hurt |
Closest papers by meaning
- Embracing Language Inclusivity and Diversity in CLIP through Continual Language Learning (AAAI 2024 · similarity 0.90)
- CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer (AAAI 2024 · similarity 0.90)
- LRM-LLaVA: Overcoming the Modality Gap of Multilingual Large Language-Vision Model for Low-Resource Languages (AAAI 2025 · similarity 0.90)
- Dynamic Adapter with Semantics Disentangling for Cross-lingual Cross-modal Retrieval (AAAI 2025 · similarity 0.89)
- Multilingual Diversity Improves Vision-Language Representations (NeurIPS 2024 · similarity 0.89)
- Large Multilingual Models Pivot Zero-Shot Multimodal Learning across Languages (ICLR 2024 · 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.
