ArtBench-10

ArtBench-10是用于基准图稿生成的类平衡，高质量，清晰注释，标准化数据集。该数据集包括来自10种独特艺术风格的60,000幅艺术品图像，以及每种风格的5,000幅训练图像和1,000幅测试图像。
与以前的艺术品数据集相比，ArtBench-10有几个优点:
首先，它是类平衡的，而大多数以前的艺术品数据集都存在长尾类分布的问题。第二，图像质量高，注释干净。第三，通过标准化的数据收集，注释，过滤和预处理程序创建了ArtBench-10。
作者以不同的分辨率 (32 × 32、256 × 256和原始图像大小) 提供了三个版本的数据集，其格式易于被流行的机器学习框架采用。

ArtBench-10 is a class-balanced, high-quality, meticulously annotated and standardized dataset for benchmarking artwork generation. This dataset contains 60,000 artwork images across 10 distinct artistic styles, with 5,000 training images and 1,000 test images for each style.
Compared with previous artwork datasets, ArtBench-10 has several prominent advantages: First, it is class-balanced, whereas most existing artwork datasets suffer from long-tailed class distribution. Second, it features high-quality images with clean annotations. Third, ArtBench-10 is constructed via standardized data collection, annotation, filtering and preprocessing workflows.
The authors provide three variants of the dataset at different resolutions (32 × 32, 256 × 256, and original image size), with formats compatible with popular machine learning frameworks.

AI搜集汇总

背景与挑战

背景概述

ArtBench-10是一个用于基准图稿生成的标准化数据集，包含10种艺术风格的60,000幅高质量图像，具有类平衡和清晰注释的特点，解决了以往艺术品数据集的长尾分布问题。数据集提供三种分辨率版本，格式易于机器学习框架使用，旨在支持艺术生成模型的评估和开发。