| # Urban3D Dataset (Sample) | |
| This is a sample subset of the Urban3D dataset, demonstrating the structure and format of our large-scale collection of real-world, object-centric multiview videos and 3D reconstructions for advancing machine learning research in autonomous systems, smart infrastructure, and urban scene understanding. | |
| ## Dataset Overview | |
| This sample includes select examples from our dataset which contains: | |
| - 10,000+ multiview videos across urban categories | |
| - Categories include: traffic signage, dumpsters, construction equipment, e-scooters, road hazards, and more | |
| - Each object derived from single video sequences | |
| - Processed through custom COLMAP-based pipeline | |
| ## Technical Details | |
| Each dataset entry includes: | |
| - High-resolution RGB frames | |
| - COLMAP-derived camera poses | |
| - COLMAP database with feature matches | |
| - Sparse 3D point clouds | |
| - Distortion-aware camera intrinsics and extrinsics | |
| - NeRF-ready transforms.json | |
| - Original video file for traceability | |
| ## Pipeline Features | |
| The dataset processing pipeline supports: | |
| - Structure-from-Motion (SfM) via COLMAP | |
| - Optimized SIFT feature extraction | |
| - Optional image undistortion | |
| - Gaussian Splatting compatibility | |
| - Clean, calibrated geometry for: | |
| - NeRF training | |
| - Gaussian splat generation | |
| - Simulation workflows | |
| ## License | |
| MIT License | |
| ## Contact | |
| For questions or support, please contact team@zeroframe.ai | |