RGBD20K Dataset Provides 160 Fine-Grained Semantic Categories for Segmentation
September 23, 2026
RGBD20K introduces a large-scale RGB-D semantic segmentation benchmark containing 20,000 image pairs. It expands the available semantic space to 160 categories, significantly exceeding the class density of NYUv2 and SUN RGB-D.
HOW THIS AFFECTS YOU
●
builderYou can leverage this larger dataset to improve the robustness of vision-based robotics or AR applications.
●
researcherYou can use this to train more generalizable segmentation models across broader semantic domains.