OnePose & OnePose-LowTexture Datasets

The OnePose dataset contains over 450 video sequences of 150 objects. For each object, multiple video recordings, accompanied camera poses and 3D bounding box annotations are provided. These sequences are collected under different background environments, and each has an average recording-length of 30 seconds covering all views of the object. The dataset is randomly divided into training and validation sets. For each object in the validation set, we assign one mapping sequence for building the SfM map and use a test sequence for the evaluation. Additionally, we construct a complementary test set, referred to as OnePose-LowTexture, specifically designed to evaluate the performance of one-shot pose estimators in challenging textureless scenarios. The OnePose-LowTexture test set consists of 40 low-textured household objects and 80 sequences. For each object, two videos are captured with different backgrounds: one serving as the reference video and the other as the test video.