MappingSeagrass/README.md

SeagrassMeadowsMapping

Description

End to end machine learning project for Mapping Seagrass Meadows with Satellite Imagery and Computer Vision.

Set Up

Recommmended to create an independent environment for the project and install the necessary libraries with provided requirements file :

conda create --name <my-env>
pip install -r requirements.txt

Architecture

For the end to end pipeline please observe the following diagram :

Data

Refer to the data DVC folder and connected task 6 folders

Modeling

Modeling was done following different methods :

• Classic machinelearning : SVVM, Random Forest , etc ...
• Deep Learning : Unet , etc ... Some training experiments and runs are tracked with MLflow in this Dasghub repository. Corresponding models are registred in MLflow registry and are for serving to REST apis.

Deployment

A web app is implemented to do inference with the most accurate model from the registry. The user can upload one or many tiff files and dowload the segmented seagrass meadow map. The web app is multi page and is implemented with dash libray. The first page is a graphical user interface for inference. The second page is a summary of the model metrics and parameters used for inference.

To launch simply run the following from command line :

python my_app.py