Research map / Vision-language and multimodal / Human and embodied modeling
"“This Is My Unicorn, Fluffy”: Personalizing Frozen Vision-Language Representations"
Niv Cohen, Rinon Gal, Eli A. Meirom, Gal Chechik, Yuval Atzmon
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 |
|---|---|---|---|
| learn word embeddings in CLIP input space (PALAVRA) instead of AvgIm+Text | CLIP · DeepFashion2 retrieval, rich query | MRR: 18.8 -> 28.4 | Improved |
| learn word embeddings in CLIP input space (PALAVRA) instead of AvgIm+Text | CLIP · YTVOS retrieval, rich query | MRR: 47.2 -> 61.2 | Improved |
| use adapter on CLIP text output (COLLIE) with rich queries rather than concept-only | COLLIE · DeepFashion2 retrieval | MRR: 13.8 -> 7.9 | Hurt |
| add test-time tuning of the concept code (ablation) | PALAVRA · DeepFashion2 retrieval | MRR: 22.1 -> 28.4 | Improved |
| add test-time tuning of the concept code (ablation) | PALAVRA · YTVOS retrieval | MRR: 47.1 -> 61.2 | Improved |
| add text augmentation with extended vocabulary (ablation) | f_theta (no tuning) · DeepFashion2 retrieval test | MRR: 19.1 -> 22.1 | Improved |
| add text augmentation with extended vocabulary (ablation) | f_theta (no tuning) · YTVOS retrieval test | MRR: 44.4 -> 47.1 | Improved |
| add cycle loss to ground-truth regularization (ablation) | f_theta (no tuning) · DeepFashion2 retrieval test | MRR: 19.2 -> 22.1 | Improved |
| add cycle loss to ground-truth regularization (ablation) | f_theta (no tuning) · YTVOS retrieval test | MRR: 41.4 -> 47.1 | Improved |
| add ground-truth regularization to cycle loss (ablation) | f_theta (no tuning) · DeepFashion2 retrieval test | MRR: 16.1 -> 22.1 | Improved |
| predict initial code with f_theta before tuning instead of random init (ablation) | PALAVRA · YTVOS retrieval test | MRR: 57.8 -> 61.2 | Improved |
| predict initial code with f_theta before tuning instead of random init (ablation) | PALAVRA · DeepFashion2 retrieval test | MRR: 27.5 -> 28.4 | Improved |
Closest papers by meaning
- Improving Personalized Search with Regularized Low-Rank Parameter Updates (CVPR 2025 · similarity 0.90)
- Ego: Embedding-Guided Personalization of Vision-Language Models (CVPR 2026 · similarity 0.89)
- Contextualized Visual Personalization in Vision-Language Models (ICML 2026 · similarity 0.89)
- MM-R1: Unleashing the Power of Unified Multimodal Large Language Models for Personalized Image Generation (AAAI 2026 · similarity 0.89)
- ConCon-Chi: Concept-Context Chimera Benchmark for Personalized Vision-Language Tasks (CVPR 2024 · similarity 0.88)
- Yo'Chameleon: Personalized Vision and Language Generation (CVPR 2025 · similarity 0.88)
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.
