New TPIPS metric captures nuanced visual similarity, outperforming existing models

Researchers have introduced a new metric called Text-Prompted Image Perceptual Similarity (TPIPS) designed to capture nuanced aspects of visual similarity, unlike existing metrics that provide a single scalar value. This metric is trained on a large dataset of human similarity judgments across image triplets, annotated with free-form semantic aspects. Benchmarking various vision-language models showed a performance gap compared to human annotators, which TPIPS aims to bridge. The TPIPS metric aligns more closely with human perception and enables new applications in text-guided retrieval and generative model evaluation. AI

IMPACT This new metric could improve the accuracy and flexibility of image retrieval and generative model evaluation systems.

RANK_REASON The cluster describes a new academic paper introducing a novel metric for image similarity. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• ScienceCast
• TPIPS

• paper
• model release

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