ETH3D

ETH3D is a benchmark and dataset for 3D reconstruction and computer vision, developed by the Computer Vision Laboratory at ETH Zurich and collaborating researchers. It is designed to support the evaluation of multi-view stereo, depth estimation, and related 3D reconstruction tasks by providing real-world data with high-quality ground truth.

The collection comprises multiple scenes organized into indoor and outdoor subsets. For each scene, ETH3D provides calibrated color images, camera intrinsics and extrinsics, and dense ground-truth depth maps or 3D point clouds derived from precise 3D scans. Some sequences include challenging material such as reflective or textureless surfaces and varying lighting.

An accompanying evaluation framework standardizes performance reporting. Metrics commonly reported include accuracy and completeness of reconstructed geometry with respect to the ground truth, computed on a per-scene basis. The dataset ships with scripts and guidelines to facilitate reproducible benchmarking across methods.

ETH3D is publicly available for research use through the project website. It is widely used to compare multi-view stereo and SLAM pipelines, informing algorithm development and enabling fair, cross-study comparisons.

The dataset has contributed to establishing common benchmarks in 3D reconstruction and has influenced subsequent datasets and competitions in the field.