Research map / Vision-language and multimodal / Cross-modal learning
Distributional Vision-Language Alignment by Cauchy-Schwarz Divergence
Wenzhe Yin, Zehao Xiao, Pan Zhou, Shujian Yu, Jiayi Shen, Jan-Jakob Sonke, Stratis Gavves
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 CS divergence to InfoNCE alignment (vs Eclipse L2 prior) | adapter + Kandinsky decoder · MSCOCO T2I | FID: 16.53 -> 12.62 | Improved |
| add CS divergence (vs Eclipse) | adapter + Karlo decoder · MSCOCO T2I | FID: 23.67 -> 11.27 | Improved |
| train alignment on CC3M with CS divergence | adapter · MSCOCO T2I | FID: 26.73 -> 22.88 | Improved |
| train alignment on CC12M with CS divergence | adapter · MSCOCO T2I | FID: 26.98 -> 22.72 | Improved |
| train alignment on LAION-HighRes 5M with CS divergence | adapter · MSCOCO T2I | FID: 19.16 -> 14.79 | Improved |
| use LoRA instead of adapter (ablation) | CS-Aligner Kandinsky · MSCOCO T2I | FID: 12.62 -> 13.52 | Mixed |
| use LoRA instead of adapter (ablation) | CS-Aligner Karlo · MSCOCO T2I | FID: 11.27 -> 15.63 | Hurt |
| add 40K paired plus 80K unpaired samples for CS divergence (ablation) | CS-Aligner Kandinsky · MSCOCO T2I | FID | Improved |
| use multiple captions per image in CS divergence (ablation) | CS-Aligner · MSCOCO T2I | FID | Improved |
| add token-level alignment (ablation) | CS-Aligner Kandinsky · MSCOCO T2I | FID | Improved |
| replace CS divergence with KL divergence (ablation) | CS-Aligner Kandinsky · MSCOCO T2I | FID: 23.48 -> 12.62 | Improved |
| replace CS divergence with Wasserstein distance (ablation) | CS-Aligner Kandinsky · MSCOCO T2I | FID: 18.41 -> 12.62 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CLIP-Adapter — CLIP-Adapter: Better Vision-Language Models with Feature Adapters (IJCV 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Linguistic-Aware Patch Slimming Framework for Fine-grained Cross-Modal Alignment (CVPR 2024 · similarity 0.91)
- Do Vision and Language Encoders Represent the World Similarly? (CVPR 2024 · similarity 0.89)
- Seeing the Image: Prioritizing Visual Correlation by Contrastive Alignment (NeurIPS 2024 · similarity 0.89)
- Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models (NeurIPS 2025 · similarity 0.89)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.89)
- AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding (NeurIPS 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.
