ETH3D Dataset

Introducing a novel multi-view stereo benchmark designed to overcome the limitations of existing datasets. This dataset features a wide range of indoor and outdoor scenes, captured with a high-precision laser scanner, high-resolution DSLR cameras, and synchronized low-resolution stereo videos. With a focus on diverse viewpoints and scene types, it is the perfect tool for advancing stereo vision techniques.

Key Features:

• High-Resolution DSLR Imagery:
  Offers high-quality images for precise depth estimation, ideal for evaluating stereo vision methods and depth perception algorithms.

• Synchronized Low-Resolution Stereo Videos:
  Captured from multiple perspectives and with varying fields of view, these videos are essential for testing real-time multi-view stereo systems.

• Diverse Scene Types:
  Includes both natural scenes and man-made environments (indoor and outdoor), ensuring a wide range of test cases for object reconstruction and 3D modeling.

• Hand-Held Mobile Device Use Case:
  The first dataset of its kind to provide data focused on hand-held mobile devices, offering higher practical relevance for mobile stereo vision applications.

• Laser Scan Alignment:
  Images are precisely aligned with the laser scans using a novel technique that minimizes photometric errors based on scene geometry, ensuring the dataset’s accuracy and reliability.

• Higher Temporal and Spatial Resolution:
  This dataset provides high temporal and spatial resolution, a significant step up from previous benchmarks like Strecha and KITTI, enabling more detailed and accurate model evaluations.

Potential Applications:

• Stereo Vision and Depth Estimation:
  Ideal for testing and developing depth estimation algorithms in both stationary and mobile settings.

• Multi-View Stereo Methods:
  Perfect for enhancing multi-view stereo methods used in 3D reconstruction and computer vision research.

• Mobile Device Integration:
  Designed with the growing use of hand-held mobile devices in mind, enabling stereo vision applications on smartphones and other mobile platforms.