Minimal Texture Dataset for RGB-D SLAM
The Minimal Texture Dataset for RGB-SLAM consists of 16 challenging sequences for RGB-D SLAM in textureless environments
Details
The Minimal Texture Dataset for RGB-SLAM consists of 16 conceptually simple but challenging sequences. We group the sequences as Extreme Geometry, Loop, Sand, and Easy. The Easy set contains control sequences to give an indicative measure of accuracy. Extreme Geometry sequences form the core of the dataset, focusing on minimal geometric content and strong perspective changes. The Loop set alternates between conceptual geometry content and the laboratory environment. Finally, the Sand group is meant to test semi-direct SLAM algorithms in textureless scenarios such as with planetary exploration purposes. The dataset was recorded with a Realsense D435i, capturing intensity and depth images of resolution 1920×1080 at 30 Hz. We used a ceiling-mounted Vicon system to record millimeter-level ground truth for the camera pose.
Link for downloadː https://zenodo.org/records/10453700
Publications
Alejandro Fontan, Riccardo Guibilato, Laura Oliva, Javier Civera, Rudolph Triebel, "SID-SLAM: Semi-Direct Information-Driven RGB-D SLAM" in Proc of the IEEE - Institute of Electrical and Electronics Engineers, pp. 1-8, 2023. [elib]