• Add DJI M2ED drone model.
• Add colour, depth, and silhouettes from an exocentic view.
• Add colour, and depth, from an egocentric view.
• Add UAVA toolset.
• Add data loaders.
• Add normals, and optical flow from an exocentic view.
• Add normals, and optical flow from an egocentic view.
• Add extra drone models.

A set of tools for working with the UAVA dataset

• PyTorch data loaders
• Dataset splits
• Visualisation script

The UAVA dataset was generated following a carefully data synthesis pipeline for ensuring the generation of photorealistic images and realistic trajectories.

Please download the trajectories before using the scripts.

An example data loading usage can be found in 'visualize_dataset.py' where the dataset is loaded and visualized using visdom.

Given that UAVA dataset can be used in a variatety of tasks ranging from computer vision to robotics, data loading can be modified accordingly w.r.t. drone model , camera views, image types and time frames.

Apart from the default values, there are some system specific arguments that need to be set up. We provide two indicative cases about how the data structure of each folder should look like.

• root_path that points the root folder of the data (e.g. "D:\\Data\\root").

• trajectory_path that points to the trajectory file (e.g. "D:\\Data\\trajectories").

The data splits follow the same distribution of the Matterport3D dataset and can be found in the data splits folder.