重新思考专业设计数据的预训练：JONES-19 数据集的证据
原标题：Rethinking Pretraining for Specialized Design Data: Evidence from the JONES-19 Cultural Design Dataset
AI 摘要
arXiv 上发布了一篇关于设计数据预训练的研究论文，基于 JONES-19 文化设计数据集，比较了 ImageNet 预训练与从零开始训练 CNN 的性能。研究发现，虽然 ImageNet 预训练能提升判别性能，但通过多裁剪局部采样从零开始训练也能达到类似效果，表明在专业设计领域，精心策划的小型高质量数据集可能比大规模通用预训练更有效。
正文节选
Computer Science > Machine Learning Title:Rethinking Pretraining for Specialized Design Data: Evidence from the JONES-19 Cultural Design Dataset View PDF Abstract:Design and architectural archives encode expert human knowledge in graphical formats, providing a critical testbed for design-inspired Machine Learning (ML) challenges absent with typical computer vision benchmarks. Building on JONES-19, a small-size image dataset based on The Grammar of Ornament (London, 1857), we evaluate