Research map / Vision-language and multimodal / CLIP and zero-shot
CLIP-like Model as a Foundational Density Ratio Estimator
Fumiya Uchiyama, Rintaro Yanagi, Shohei Taniguchi, Shota Takashiro, Masahiro Suzuki, Hirokatsu Kataoka, Yusuke Iwasawa, Yutaka Matsuo
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 |
|---|---|---|---|
| importance-weight CLIP pretraining loss with CLIP-based density ratio from a single domain prompt | CLIP pretraining on CC12M · Oxford-IIIT Pet zero-shot | accuracy and F1: +2-8 points accuracy, +3-7 points F1 at epochs 4-6 | Improved |
| importance-weighted learning with domain prompts (food, pets, flowers) | CLIP pretraining on CC12M · Food101, Pets, Flowers102 zero-shot | accuracy and F1 | Improved |
| filter DataComp small pool to top 25% by text D_KL | CLIP pretraining on DataComp small · ImageNet-1k zero-shot | accuracy: 0.025 -> 0.0300 | Improved |
| filter DataComp small pool to top 25% by text D_KL | CLIP pretraining on DataComp small · mean over 38 tasks | mean metric: 0.132 -> 0.1337 | No clear effect |
| use text-based KL divergence instead of image-based KL divergence for data filtering (ablation) | D_KL filtering · DataComp small, mean over 38 tasks | mean metric: 0.122 -> 0.1337 | Improved |
| use CLIPScore filtering instead of KL-based filtering (ablation) | DataComp small filtering · DataComp small, mean over 38 tasks | mean metric: 0.1337 -> 0.173 | Improved |
| filter by image-based D_C (centered norm) instead of text-based (ablation) | DataComp small filtering · ImageNet-1k zero-shot | accuracy: 0.0312 -> 0.0108 | Hurt |
| KL-guided filtering versus LAION2B filtering | DataComp small filtering · DataComp small, mean over 38 tasks | mean metric: 0.133 vs 0.1337 | No clear effect |
Closest papers by meaning
- Demystifying CLIP Data (ICLR 2024 · similarity 0.89)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.88)
- CLIP the Bias: How Useful is Balancing Data in Multimodal Learning? (ICLR 2024 · similarity 0.88)
- CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation (NeurIPS 2025 · similarity 0.87)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.87)
- Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotations (CVPR 2024 · similarity 0.87)
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.
