A stereo-depth dataset for microsurgical scenes of soft tissues

MicroRGBD is a high-precision stereo RGB-D dataset designed for microsurgical scene understanding. It comprises 3,586 synchronised stereo image pairs (1280×960) with pixel-aligned depth ground truth and per-pixel confidence annotations, acquired by a Photoneo PhoXi XS structured-light scanner with measured depth repeatability of approximately 7 μm. The dataset spans 6 scene categories, 5 tissue types, and 25 microsurgical instrument types across 34 acquisition sessions. A motion-compensated acquisition protocol generates per-view depth maps registered to both left and right stereo cameras. Complete calibration parameters and official train/val/test splits are included. For full details, please refer to the accompanying paper.