Moonshine
Moonshine is a Python library that makes it straightforward for distant sensing researchers,
professionals, and enthusists to develop ML fashions on their information. It offers
pre-trained models throughout a wide range of
datasets
and architectures, permitting you to cut back your labeling prices and compute necessities
to your personal utility.
Why Use Moonshine?#
-
Pretrained on multispectral information: Many current packages are pretrained with
ImageNet or comparable RGB photographs. Utilizing Moonshine you possibly can unlock the complete energy of
satellites that many comprise many channels of multispectral information. -
Pretrained on distant sensing information: Pretraining within the area of your information is
necessary, and most off the shelf pretrained fashions are match to pure photographs akin to
ImageNet. -
Concentrate on usability: Whereas there are some educational distant sensing pretrained
fashions accessible, they typically are tough to make use of and lack assist. Moonshine is
designed to be straightforward to make use of and can supply group assist through Github and Slack.
Want extra convincing that Moonshine works? Try this comparability of Moonshine
pretrained weights vs coaching from scratch:
The above chart reveals the distinction between coaching the
functional map of the world classification activity utilizing
our pre-trained mannequin vs. coaching from scratch. The duty is to categorise patches of
satellite tv for pc information by the practical objective of the land, with 63 doable courses and over
300,000 coaching photographs.
Coaching from scratch each performs worse total, and for roughly the identical stage of
accuracy we will practice for 45% much less time (28h vs 16h on a V100). Try the
quick start part for additional data, together with
learn how to set up the library.