PQPP

PQPP数据集是由布加勒斯特大学创建的，专注于文本到图像的提示和查询性能预测。该数据集包含10,000个查询，用于评估图像生成和检索的难度。数据集从Microsoft Common Objects in Context (MS COCO)中选取，通过k-means聚类算法筛选出代表性提示。创建过程中，收集了超过1.5百万的人工标注，用于评估生成和检索性能。PQPP数据集旨在解决文本到图像生成和检索中的性能预测问题，为研究者提供了一个评估和比较不同模型性能的平台。

PQPP Dataset was created by the University of Bucharest, focusing on text-to-image prompt and query performance prediction. The dataset includes 10,000 queries designed to evaluate the difficulty of image generation and retrieval. It is sourced from Microsoft Common Objects in Context (MS COCO), where representative prompts were selected via the k-means clustering algorithm. Over 1.5 million human annotations were collected during the dataset's creation to assess generation and retrieval performance. The PQPP dataset aims to solve the performance prediction problem in text-to-image generation and retrieval, providing researchers with a platform to evaluate and compare the performance of different models.