10+ depth datasets across indoor and outdoor categories for training generalizable monocular depth estimation models. It includes millions of RGB-depth pairs sourced from environments like NYU Depth V2 and KITTI.
Use Cases
- Train monocular depth estimation models using the RGB images and corresponding depth maps
- Evaluate cross-domain generalization by testing models on unseen datasets within the collection
- Benchmark zero-shot transfer performance across different camera intrinsics and scenes
Strengths
- Aggregates 10+ datasets including NYU Depth V2, KITTI, and ScanNet
- Provides millions of RGB-depth image pairs
- Standardizes data from multiple sensors including LiDAR and structured light