UAV3DCrop

Repeated multi-angle UAV surveys of corn, soybean, wheat and oat fields from 2023 to 2025, with refined camera poses and sparse point clouds for 3D reconstruction.

Project page: UAV3DCrop

Paper: UAV3DCrop: Benchmarking 3D Reconstruction in Repeated Multi-Angle UAV Crop Surveys

Depth maps and ground measurements: UAV3DCrop Depth

Structure

<year>/Day<NNN>_<Crop>/
├── images/            RGB images (1.JPG, 2.JPG, ...)
├── transforms.json    camera intrinsics and poses (nerfstudio format)
└── sparse_pc.ply      sparse point cloud

DayNNN is the day count from the first UAV flight of that year (Day001).

Coordinates

Camera positions in transforms.json and points in sparse_pc.ply are in metres, in a local frame with x east, y north and z up. Scenes that differ only in DayNNN (for example all 2025/Day*_Corn) share one frame, so they are aligned across days. Other scenes are in separate frames and are not aligned with each other.

License

CC BY 4.0; see LICENSE. Please cite the UAV3DCrop paper when using the dataset.