-
-
Logo
PlaDIS (Plant Disease Identification System)
We made this project for the Quarantine track because it will help people who have taken up gardening as a hobby during the lockdown. We used a dataset from Kaggle containing about 41,000 images of different plants' leaves with different diseases. After curating the dataset, we uploaded it to Google Firebase AutoML Vision Edge to train for 10 predicted compute hours, which ended up finishing in about 4 hours. We built an accompanying Android application that allows users to take pictures of their plants' leaves and remotely accesses the model on Firebase to classify potential diseases the leaves might have.
We were both quite new to Android development, so we had the opportunity to learn how to use the Android's libraries for the camera and how to integrate the application with Google Cloud Platform and Firebase. We also learned how to be efficient with our time. While the model was training, we managed to build the entire camera functionality for the application.
We had a lot of bugs with the camera functionality, especially because the documentation is not particularly readable or accessible for the Camera2 library in Android. The examples were also far and few in between, so we had to perform a lot of experimentation to get it to work.
We also believe that our model may be slightly overtrained as we noticed that the test and validation precision values were rather high (very close to 100%). We did not have the opportunity to re-train the model as it would take too long and we were dealing with other bugs related to the Android application. If we had more time, we would go back and re-train the model so that it isn't overtrained and biased towards the training data.
Log in or sign up for Devpost to join the conversation.