Stereo Correspondence and Reconstruction of Endoscopic Data Sub-Challenge¶

Part of the Endoscopic Vision Challenge¶

This challenge will focus on depth estimation from endoscopic data. Depth estimation and scene reconstruction are important components of many endoscopic navigation and augmented reality guidance system. Endoscopic images can differ considerably from natural scenes typically used in computer vision. In particular, the harsh, directional lighting, smooth, non-planar surfaces and subtle texture can pose challenges to traditional methods. Previous open datasets for endoscopic data have used isolated ex-vivo or phantom organs with registered models obtained from CT or structured light. However, these techniques limit the depth estimation problem to artificially isolated organs and do not provide direct measurement of the ground-truth stereo correspondences. As a result, errors from registration, camera-calibration and CT segmentation are not separated from the stereo-correspondence.

With this dataset, we hope to provide a similar baseline to the successful Middlebury Stereo Dataset which has been successful in advancing the state of the art in stereo methods over the last 2 decades.

Prizes¶

Intuitive Surgical will be offering a prize of \$4000 for the winner of this challenge and \$2000 for the runners up.