HPatches
收藏资源简介:
HPatches数据集是由帝国理工学院创建的一个大型图像序列数据集,用于评估和训练现代局部图像描述符。该数据集包含多种场景和捕捉条件,如视角、光照和时间变化图像序列,共计约105000条数据。数据集的创建过程涉及使用多种尺度不变兴趣点检测器提取特征,并根据地面真实变换确定唯一对应关系。HPatches数据集主要应用于图像匹配、检索和分类等任务,旨在解决现有数据集在评估局部描述符性能时的不足和不一致性问题。
The HPatches dataset is a large-scale image sequence dataset created by Imperial College London, designed for evaluating and training modern local image descriptors. It encompasses diverse scenarios and capture conditions, including image sequences with varying viewpoints, illuminations and temporal changes, totaling approximately 105,000 samples. The dataset construction process involves extracting features using multiple scale-invariant interest point detectors, and establishing unique correspondences based on ground-truth transformations. The HPatches dataset is primarily applied to tasks such as image matching, retrieval and classification, aiming to address the shortcomings and inconsistencies of existing datasets when evaluating the performance of local descriptors.




