Leandrone
The "Leandrone" datasets are collections of images mimicking the photo taken by the front camera of a Bitcraze AI deck 1.1 mounted on a Crazyflie 2.1 nanodrone. The photos are monochrome and have a specific resolution. The aim of these data sets is to create a deep learning model for an autonomous drone that follows a line, according to certain specifications (they are different for V1 and V2). The data was generated using the 3D modeling application "Blender".
"LEANDRONE_V1" is a collection of 500 images with labels. Each image has a label with 2 points named "from" and "to" that draw a line in the forward direction. Each image has random variations in camera angles, line trajectory, location of lights and furniture.
"LEANDRONE_V2" is a collection of 240 images with labels. Each image has a label indicating the angle between the camera and the line path. This indicate where the drone should turn. Each image has random variations in the line layout, the number of bottles and the positioning of the bottles around the course (bottle counting was an optional objective). A 3D model of the course was reconstructed using photogrammetry.
More informations in the "README" files of the Huggingface repositories.
