AstroCompress
收藏资源简介:
AstroCompress数据集由加州大学尔湾分校、加州大学伯克利分校和劳伦斯伯克利国家实验室创建,旨在解决天文观测数据传输瓶颈问题。该数据集包含5个不同的数据集,涵盖了多种观测条件、探测器技术和动态范围,总数据量约为320GB。数据集包括地面和太空观测的16位无符号整数成像数据,以及多波长和时间序列成像数据。该数据集的设计目的是为了促进机器学习社区对天文物理数据压缩的研究,并评估了7种无损压缩方法的性能,展示了神经压缩在科学应用中的巨大潜力。
The AstroCompress dataset was developed by the University of California, Irvine, the University of California, Berkeley, and Lawrence Berkeley National Laboratory, aiming to address the bottleneck problem in astronomical observation data transmission. The dataset consists of five distinct subsets covering a wide range of observation conditions, detector technologies, and dynamic ranges, with a total data volume of approximately 320 GB. It includes 16-bit unsigned integer imaging data from both ground-based and space-based observations, as well as multi-wavelength and time-series imaging data. This dataset is designed to promote research on astrophysical data compression within the machine learning community, and it evaluates the performance of seven lossless compression methods, demonstrating the considerable potential of neural compression in scientific applications.

- 1AstroCompress: A benchmark dataset for multi-purpose compression of astronomical data加州大学尔湾分校;加州大学伯克利分校;劳伦斯伯克利国家实验室 · 2025年



