Research map / Vision-language and multimodal / Cross-modal learning
Towards Cross-modal Backward-compatible Representation Learning for Vision-Language Models
Young Kyun Jang, Ser-nam Lim
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 text-only pretrained projection module phi (XBT vs Base trained directly) (ablation) | Base (LoRA + soft prompt) · nocaps (B32 old / L14 new) new query vs old gallery | R@1 text query to image gallery: 43.48 -> 48.02 | Improved |
| add text-only pretrained projection module phi (XBT vs Base trained directly) (ablation) | Base (LoRA + soft prompt) · nocaps (B32 old / L14 new) new query vs old gallery | R@1 image query to text gallery: 69.78 -> 75.02 | Improved |
| use XBT instead of full fine-tuning (ablation) | Full-tune · nocaps (B32 old / L14 new) new query vs old gallery | R@1 image query to text gallery: 63.76 -> 75.02 | Improved |
| add soft prompt on top of LoRA-only (ablation) | LoRA-only · nocaps (B32 old / L14 new) new query vs old gallery | R@1 text query to image gallery: 43.40 -> 43.48 | Improved |
| add Gaussian noise during text-only pretraining (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 text query to image gallery: 47.24 -> 48.02 | Improved |
| add Gaussian noise during text-only pretraining (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 image query to text gallery: 73.31 -> 75.02 | Improved |
| scale up the text corpus D_T from 0.25x to 1x (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 text query to image gallery: 46.89 -> 48.02 | Improved |
| scale up the text corpus D_T from 0.5x to 1x (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 image query to text gallery: 73.89 -> 75.02 | Improved |
| double the supervised image-text pairs (2x D) (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 text query to image gallery: 48.02 -> 47.63 | No clear effect |
| halve the supervised image-text pairs (0.5x D) (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 image query to text gallery: 75.02 -> 71.69 | Hurt |
| use text-only pretraining instead of image-only pretraining (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 text query to image gallery: 45.32 -> 48.02 | Improved |
| replace supervised data with CC3M (ablation) | XBT · nocaps (B32 old / L14 new) | R@1 image query to text gallery: 75.02 -> 71.33 | Hurt |
Closest papers by meaning
- Dynamic Adapter with Semantics Disentangling for Cross-lingual Cross-modal Retrieval (AAAI 2025 · similarity 0.91)
- CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer (AAAI 2024 · similarity 0.90)
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- UniAdapter: Unified Parameter-Efficient Transfer Learning for Cross-modal Modeling (ICLR 2024 · similarity 0.90)
- Adaptive Prompt-Based Semantic Embedding with Inspire Potential of Implicit Knowledge for Cross-Modal Retrieval (AAAI 2025 · similarity 0.90)
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.90)
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.
