-
app_with_data
-
-
-
-
-
-
Inspiration
Phytoplankton, such as Phaeocystis antarctica, occupy a key niche in the Antarctic food web and help regulate global climate by pulling carbon dioxide from the atmosphere to fuel their own explosive growth. A better understanding of why P. antarctica is so successful could help scientists predict its future distribution as global warming continues and ice cover shrinks in the Southern Ocean. It then becomes crucial for the scientific teams to have an user-friendly and reliable tool to classify their samples. Current state of the art algorithm used by research groups perform poorly (low accuracy), is slow and not easily accessible.
What it does
This project consists in two parts:
- a deep learning algorithm for classification, using keras and tensorflow.
- a web application where the users can upload data, predict the species of phytoplankton and download the results (some samples are available here for testing: https://github.com/thomashossler/phytoplankton-classification/tree/master/samples)
How I built it
I built the backend in python (flask app) and the front end in angular. My approach started with a focus on the algorithm itself, followed by the concern of making the algorithm accessible and user friendly.
Challenges I ran into
My frontend skills are close to zero. I am not familiar with angular and have very little knowledge in css and html. Deploying the web app on AWS was also quite challenging for me. Finally, the algorithm tended to overfit.
Accomplishments that I'm proud of
Creating the web application, fast, simple and easy to use.
What I learned
A lot of frontend skills. A few ML skills. It's one thing to create a reliable algorithm. It's another to deploy it and make it accessible to the user.
What's next
A few ideas:
- the ability for the user to upload massive amount of data and retrain the algorithm.
- connect the backend with a database
- a even better UI
Log in or sign up for Devpost to join the conversation.