LMR
收藏资源简介:
LMR是由中国科学院自动化研究所和百度公司联合创建的大规模多参考图像超分辨率数据集。该数据集包含112,142组300×300像素的训练图像,每组图像配有5张不同相似度的参考图像,旨在解决现有单参考图像训练与实际应用中多参考图像测试之间的不匹配问题。LMR数据集的创建过程涉及从MegaDepth数据集中筛选相似图像对,并通过特定的预处理步骤确保图像对的适用性。该数据集主要应用于图像超分辨率技术,特别是在需要多参考图像的场景中,以提升图像恢复的质量和细节表现。
LMR is a large-scale multi-reference image super-resolution dataset jointly developed by the Institute of Automation, Chinese Academy of Sciences, and Baidu Inc. This dataset comprises 112,142 sets of 300×300 training images, where each set is accompanied by 5 reference images with different similarity levels. It is designed to address the mismatch between single-reference image training in existing research and multi-reference image testing in real-world applications. The construction of the LMR dataset includes screening similar image pairs from the MegaDepth dataset and performing specific preprocessing procedures to guarantee the suitability of these image pairs. This dataset is primarily utilized for image super-resolution techniques, particularly in scenarios requiring multi-reference images, to improve the quality and detail restoration performance of reconstructed images.

- 1LMR: A Large-Scale Multi-Reference Dataset for Reference-based Super-Resolution中国科学院自动化研究所 · 2023年



