New TLP framework adapts vision-language models without source data

Researchers have introduced Test-Time Logit Prompting (TLP), a novel framework designed to improve the performance of vision-language models (VLMs) when faced with missing modalities during deployment. Unlike previous methods that require access to original training data, TLP operates in a source-free manner, adapting VLMs at test time. The framework optimizes logit prompts using uncertainty-aware adjustments and modality-complete consistency regularization to maintain prediction confidence and semantic coherence. Experiments show TLP can boost recognition performance by up to 8% with minimal tunable parameters and a few optimization steps. AI

IMPACT Enables more robust deployment of vision-language models in scenarios with limited or unavailable training data.

RANK_REASON The item describes a new research paper published on arXiv detailing a novel adaptation framework for vision-language models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• ScienceCast
• Test-Time Logit Prompting
• TLP
• vision-language model

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →