We discussed building a team for building a team at our local water hole.
What it does
Our project takes Landsat images, processed them into labeled "chips" and trains a fully connected CNN.
How we built it
We built Geopher within 4 weeks using Python and Cloud resources.
Challenges we ran into
Chipping the images and loading them in batches for training proved harder than anticipated.
Accomplishments that we're proud of
Creating a reusable pipeline, getting segmentation results and transforming them back to coordinate space.
What we learned
We learned how sensitive AI models are to data. We learned about FCCNN architecture.
What's next for Geopher
There is much more experimentation possible. See slide deck.