Browse » Aerial » Other

Top Other Datasets

There are many uses for computer vision in aerial imagery that don't quite fit in a neat bucket. Here are some examples to spark the imagination:

Rocket Detect is the dataset used to train the nueral network behind Autotrack, NASASpaceflight's automated rocket tracking system. Rocket Detect uses three classes:

  1. Engine Flames - The fire produced by the rocket
  2. Rocket Body - The body of the launch vehicle
  3. Space - A tiny spec in the sky that is the rocket after it has ascended into space

Using YoloV5 and Strongsort, Rocket Detect in its current form has proven sufficient to reliably track a Falcon 9 first stage continuously form launch to landing. It also works on a wide variety of other launch vehicles. The dataset will increase with time to further improve reliablity.

A public dataset of craters on Mars & Lunar surface.

Please notice that currently all data are from publicly published sources, a detailed data source list will be available soon.

Manual labeling has been performed for object detection purposes. For other purposes such as semantic segmentation, please first label the images in accordance with your need.

Maintained by @Lincoln-Zhou,

There is a lack of publicly available dataset for detecting people from a UAV persepective especially when they might be hidden under tree canopy. This thermal dataset (with some RGB images sprinkled) recorded using FLIR Vue TZ20 from an overhead/UAV perspective hopes to be useful in search and rescue missions.

Project Title : Real time video processing for Hazard Detection.

Dataset : Lunar Terrain

Hazards to annotate : Boulder, Crater and Plain Surface.

For annotation a video of 31s was uploaded with 45fps as frame rate to extract the frames. Out of total 924 frames extracted, all frames are done with annotation.

Dataset 2.0:

Total 3122 images extracted from a video of 3:14 minutes, divided with 1040 images each.

Contributers :

  • Ruchi Agarwal
  • Neha Bomble
  • Gayatri Deshmane

Real-time Video Processing for Hazard Detection using Deep Learning for Lunar Dataset.

Under the guidance of Dr. Dipti Patil.

Total Frames extracted: 5579.
Classes to be detected: Boulder, Crater and Plain surface.
Current Status: 391 Frames annotated.

Links to External Files Needed:

1. Hazard Detection Python Code File

2. Altitude Estimation coordinates

Labeller: Ruchi Agarwal, Neha Bomble and Gayatri Deshmane.