Research map / Vision-language and multimodal / CLIP and zero-shot
Discovering Unwritten Visual Classifiers with Large Language Models
Mia Chiquier, Utkarsh Mall, Carl Vondrick
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 |
|---|---|---|---|
| use 10 past attribute sets in the LLM prompt (in-context learning) instead of 1 (ablation) | LLM-Mutate · iNaturalist Lichen | accuracy: 31.6 -> 46.67 | Improved |
| use 10 past attribute sets in the LLM prompt instead of 1 (ablation) | LLM-Mutate · KikiBouba KB1 | accuracy: 50.3 -> 78.9 | Improved |
| use 10 past attribute sets in the LLM prompt instead of 1 (ablation) | LLM-Mutate · iNaturalist Wrasse | accuracy: 24.0 -> 42.0 | Improved |
| replace LLM evolutionary search with gradient-based token optimization | LLM-Mutate · iNaturalist Wild Rye | accuracy: 52.0 -> 40.0 | Hurt |
| learned attributes without class names versus class-name-based baselines (Class Name, Classification by Description) | CLIP zero-shot classification · iNaturalist (5 families) | accuracy: outperforms baselines by 18.4% on average | Improved |
| learned attributes versus baselines on novel concepts | CLIP zero-shot classification · KikiBouba | accuracy: average +22.2% | Improved |
| hierarchical coarse-to-fine classification with coarse-level choice of method | LLM-Mutate · iNaturalist 26-class hierarchical | accuracy | No clear effect |
| use learned attributes versus classification by description for prediction confidence | LLM-Mutate · all datasets | mean margin | Improved |
Closest papers by meaning
- Follow-Up Differential Descriptions: Language Models Resolve Ambiguities for Image Classification (ICLR 2024 · similarity 0.89)
- ProAPO: Progressively Automatic Prompt Optimization for Visual Classification (CVPR 2025 · similarity 0.89)
- The Neglected Tails in Vision-Language Models (CVPR 2024 · similarity 0.89)
- Unified Generative and Discriminative Training for Multi-modal Large Language Models (NeurIPS 2024 · similarity 0.89)
- Analyzing and Boosting the Power of Fine-Grained Visual Recognition for Multi-modal Large Language Models (ICLR 2025 · similarity 0.89)
- Tree of Attributes Prompt Learning for Vision-Language Models (ICLR 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.
