ByteDance-Seed/MagicBench

Magic Bench是一个用于评估文本到图像生成模型性能的综合性评价数据集，包含377个精心挑选并带有详细注释的提示，涵盖了多种应用场景，并提供了中英文两种语言的提示文本，以便进行跨语言评估。Magic Bench is a comprehensive evaluation dataset for assessing the performance of text-to-image generation models, containing 377 carefully curated prompts with detailed annotations across multiple dimensions, covering various application scenarios, and providing prompts in both Chinese and English for cross-lingual evaluation.

搜集汇总

背景与挑战

背景概述

MagicBench是一个用于评估文本到图像生成模型的综合数据集，包含377个带有详细注释的中英文提示，覆盖多种应用场景和表达能力，旨在全面测试模型在不同情境下的表现。