The app predicts a flower’s name. An image of flower is uploaded and sent to the server. The server randomly selects saved models and try to predict the flower’s name using each model selected. The most predicted name is returned to the user.
The saved models come from the model saved during the last project of Pytorch Udacity Challenge. The user has the possibility to upload his model. If the model is successfully uploaded, an Id is sent to the email the user has used to upload his model. When trying to predict a flower’s name, the user has either the possibility to give a model id or not. If the Id is given, only the prediction of the model with the corresponding Id on the server will be used.
Here is the live version of the app
The frontend is written in angular. A flower is uploaded to the backend and the most likely flower's name is returned. A model can be uploaded as well.
-
change directory
cd frontend -
install all the dependencies
npm install -
serve the frontend
npm start
The backend is written with the server flask. The prediction are made using pytorch.
-
change directory
cd backend -
install all the dependencies
pip -r install requirements.txt -
run the server
python -m flask run
To make predictions, the server needs to have the file checkpoint.pth saved from a flower classifier model written in pytorch
-
install docker compose if not yet installed.
-
run docker-compose
docker-compose up