KidneyDepth: A Synthetic Kidney Dataset for Metric Depth Estimation in Ureteroscopy

KidneyDepth is a dataset composed by synthetic ureteroscopic images generated by BlenderProc and real ureteroscopic images. The objectives of this dataset are to generate ground truth maps for fine-tuning and evaluating metric depth estimation models (KidneyDepthMetric), to visually evaluate them on real endoscopic kidney images (KidneyDepthfURS), and to evaluate the estimated trajectories of a SLAM system (KidneyDepthSLAM).

The approximate size of the models used can be seen in models_size.png

Changelog

Version v3:

• Fix: Depth images changed to .tif format to store the depth in float16
• Add camera information: clipping plane and FOV
• Add information about the size of the models
• Dataset reduction due to size limit of Zenodo

Version v2:

• Initial upload of the full dataset

Version 0.0.0:

• Small subset for paper review (before paper publication)