KnowVis: A Dual-View Benchmark for Diagnosing World-Knowledge Grounding in Text-to-Image Models
📖 Overview
Text-to-image (T2I) models have made substantial progress in visual realism, aesthetic quality, and instruction following. However, real-world prompts often go beyond explicit visual descriptions and require implicit facts, structured knowledge, and domain-specific commonsense. Existing evaluations mainly focus on explicit prompt-to-image semantic alignment or… See the full description on the dataset page: https://huggingface.co/datasets/QuantumWhisper42/anonymous_dataset.
