Welcome to the ETH3D Benchmark
- 13 / 12 DSLR datasets for training / testing.
- 5 / 5 multi-cam rig videos for training / testing.
- 27 / 20 frames for two-view stereo training / testing.
- High-resolution (24 Mpx) DSLR images
- Multi-camera rig videos (4 cameras, ~13.6 Hz)
This webpage presents a multi-view stereo / 3D reconstruction benchmark covering a variety of indoor and outdoor scenes.
Ground truth geometry has been obtained using a high-precision laser scanner.
A DSLR camera as well as a synchronized multi-camera rig with varying field-of-view was used to capture images.
We offer the following challenges:
- 13 training and 12 test scenes (data / results) for high-resolution multi-view stereo with images recorded by the DSLR camera
- 5 training and 5 test videos (data / results) at ca. 13.6 Hz for low-resolution many-view stereo on video data recorded with the multi-camera rig
- 27 training and 20 test frames (data / results) for low-resolution two-view stereo on frames of the multi-camera rig
- 2017-10-04: Extension with more data. Full details here.
- 2017-07-19: Initial release of the dataset.
If you use our data in your research, please cite:
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. [Bibtex][PDF][Supplementary]
The data provided here is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.