ETH3D stereo benchmark code

[Johannes Schönberger]

Dear COLMAP users,

We just released the source code for our stereo benchmark https://www.eth3d.net/ published at CVPR17:

T. Schöps, J. L. Schönberger, S. Galliani, T. Sattler, K. Schindler, M. Pollefeys, A. Geiger, "A Multi-View Stereo Benchmark with High-Resolution Images and Multi-Camera Videos", Conference on Computer Vision and Pattern Recognition (CVPR), 2017.

This might be very interesting to many of you, which is why I am sharing it here as well.

The ETH3D dataset processing tools consist of a number of programs for creating 3D reconstruction evaluation datasets from images and laser scans. This includes tools for laser scan processing (outlier removal, scan alignment, ...) and image alignment wrt. laser scans (by optimizing for color consistency among images and the scans). The tools additionally include support for semantic labeling of point clouds and limited support for scan-image alignment for depth images, which was not used for the ETH3D benchmark.

The pipeline for processing a dataset is as follows:

• Data acquisition, input is expected as image files and colored point clouds as PLY
• Creating the file system layout
• PointCloudCleaner: remove some point cloud outliers automatically
• PointCloudEditor: remove remaining point cloud outliers manually
• CubeMapRenderer: render cube map images from laser scans
• Use an external SfM tool to estimate an initial alignment between the scans and the images
• SfMScaleEstimator: estimates the scale of the SfM model
• ICPScanAligner: refine the scan alignment using point-to-plane ICP
• NormalEstimator: estimate normal vectors for the scans
• Use an external surface reconstruction tool to reconstruct a surface mesh for occlusion handling
• SplatCreator: create splats for points which are not represented in the surface mesh
• DatasetInspector: allows to view the aligned images and draw image masks
• ImageRegistrator: refine the image alignment and intrinsics using dense image alignment
• GroundTruthCreator: create the ground truth data for evaluation

More information can be found on GitHub at https://github.com/ETH3D/dataset-pipeline

Cheers,

Johannes