Developed for NASA by College and High School Students
What is Feature Hunter?
The Earth Science and Remote Sensing unit at NASA's Johnson Space Center maintains an enormous collection of over three million astronaut photographs of Earth, spanning manned spaceflight from the Mercury missions to today's ISS missions.
These photos can be vital to researchers across many fields, but many are inaccessible because they lack labels for common features like islands, volcanoes, and rivers. Feature Hunter is an important part of the creation of a machine learning algorithm aimed at labeling previously uncategorized Astronaut Photography.
Machine learning requires large sets of training data to produce accurate results. Feature Hunter users will help develop this training data by viewing images, determining whether or not a feature or features are present, and identifying features by placing bounding boxes around them.
Feature Hunter may not work in Internet Explorer. If you experience issues, ensure compatibility mode is disabled, or try Feature Hunter in another browser.
Using Feature Hunter
As a Feature Hunter user you will be asked to examine images and draw bounding boxes around features such as islands, volcanoes, and cities in those images.
This tutorial will guide you through the basics of using Feature Hunter, as well as providing useful example images.
Begin by selecting the kind of feature you want to search for by clicking on one of the images in the Select A Category section below.
Additional tutorials to help identify and draw bounding boxes around each individual kind of feature will be available once you make your selection.
After selecting a feature category an image will be displayed. Examine it closely and determine if the feature you are looking for is visible.
- If you don't see the feature you are looking for, select the No option
- If you are unsure as to whether or not the feature is present, select the Not Sure option
- If you see the feature, select the Yes option
Selecting the Yes option will bring you to a page where you will be asked to identify the feature's location by moving and resizing bounding boxes around it.
Don't hesitate to use the No and Not Sure options when appropriate - both of these options are important in creating the best training data possible.
Selecting Yes option will bring you to a page where you can move and resize bounding boxes to surround the features in your image.
If there are multiple features, or the feature you are trying to box is oddly shaped, you can add up to four more boxes using the Add Box button.
The Remove Box button will remove the newest box on the screen.
Select the Confirm Selection and confirm the correct category is selected before submitting your bounding boxes.
Submitting your bounding boxes will lock them in place - you won't be able to make changes, but you can add more boxes and make more submissions if you want to box other areas.
Selecting View New Image will display another image for you to examine.
Here are some examples of bounding boxes around a city:
Good
Box bounds majority of city and its boundaries
Okay
Box bounds city and some of the area surrounding it
Bad
Box only bounds part of city, and contains lots of the surrounding area
Good bounding boxes produce the most accurate training data. Boxes similar to the Okay example are fine, but Bad bounding boxes will result in inaccurate training data.
Select an image to take a closer look in a new tab.
Precision and accuracy are important when drawing bounding boxes, but attempt to identify every single branch of a river, or every building in a city. Instead, try to focus on the boundaries of the feature as a whole.
Clusters of features don't requre lots of individual boxes - a group of islands clustered together can be identified with one bounding box.
Some images may contain multiple features - the example below contains both a city and a river.
When this is the case, draw bounding boxes for each type of feature individually. Select the proper category for each set of bounding boxes from the dropdown menu on the confirmation dialog before submitting.
When analyzing images, you may encounter obstacles that will make identifying features and drawing bounding boxes challenging. For example:
- Cloud Cover: Cloud cover is present in a large amount of database images. If the cloud is thick enough to obscure your vision, only box what you can see. However, if you can see your feature through thin, wispy clouds, feel free to include them in your box.
- Angles: Some images will be taken at oblique angles, or will be distorted due to earth's curvature. Try your best to keep your boxes accurate, and aim to box the center of the feature if possible.
- Location of the horizon: Images may not have the horizon in a familiar position
The astronaut photographs Feature Hunter users will help in identifying will serve as valuable resources across many different scientific disciplines.
Past studies that have utilized astronaut photography include Tidal-Flat Loss in Japan and the effects of elephant populations on vegetation in Botswana. For a full list of publications and research, visit our Publication List.
