UAV-Seq3D: A Multi-Sequence UAV Dataset for Feed-Forward 3D Reconstruction

UAV-Seq3D is a multi-sequence unmanned aerial vehicle (UAV) RGB image dataset designed to support research on monocular 3D reconstruction, camera pose estimation, and UAV-view spatial perception. The dataset was collected using a DJI Mavic 3 Thermal (M3T) UAV with RTK positioning capability. It covers 13 outdoor scenes, including buildings, roads, parking areas, and farmland. For each scene, two UAV image sequences were acquired: a regular circular flight trajectory with an approximately 45-degree gimbal pitch angle, and a free or irregular flight trajectory with an approximately 60-degree gimbal pitch angle. UAV-Seq3D contains 26 RGB image sequences. Each sequence is accompanied by frame-level metadata, including local East-North-Up (ENU) coordinates and UAV attitude information, including pitch, roll, and yaw. Because of the total dataset size, UAV-Seq3D is distributed across this main Zenodo record and four linked Zenodo image-data volume records. The main record is available at https://doi.org/10.5281/zenodo.21212026. This main record provides the metadata archive, dataset documentation, package inventory, checksum file, license information, and links to all image-data volumes. The RGB image archives are provided in the four linked image-data volume records listed below. Image-data volumes: - Volume 1: https://doi.org/10.5281/zenodo.21122187- Volume 2: https://doi.org/10.5281/zenodo.21159375- Volume 3: https://doi.org/10.5281/zenodo.21188120- Volume 4: https://doi.org/10.5281/zenodo.21206046 To reconstruct the complete UAV-Seq3D dataset, users should download the metadata archive from this main record and all RGB image archives from the four linked image-data volume records. Please refer to README.md for the dataset structure, metadata fields, coordinate definition, archive extraction instructions, integrity verification procedure, license, and citation guidance.