In this project, we use computer vision to classify images of vegetables, 15 classes of common vegetables. For our traditional ML approach, we built an SVM, which had validation and test accuracies of 81.8% and 82.8%, respectively. For our naive approach, we used ResNet-18 without replacing and retraining the fully connected layer, and resulted in a validation score of 8.1% For our final approach, we used Google's Inception, which had validation and test accuracies of 99.9% and 99.8%, respectively.
s3bucket_googlecolab.py: script to grab the data from our s3 bucket and save it to your Google Drive.
SVC.ipynb: notebook for our SVC model development and tests
main_inception_VeggieTales_model.ipynb: notebook for training and evaluation of our final Inception model
restNetNaive.ipynb: notebook for the evaluation with resNet18, no training.
generateIllustration.py: script to generate an illustration using OpenAI API based on the generated story
generateStory.py: script used to call the OpenAI API to generate a child-story based on the input of detected vegetables
setup.py: script to grab data and run final Inception model.
SVC_model.py: script to run SVC model
naive_model.py: script to run naive ResNet-18 model.
main_inception_vegetable.py: script to run Inception model
files for UI
To run our application, run flask in the food-fables-backend directory.
Then in an another terminal, run npm run build and then npm start in the food-fables directory, and watch your vegetables come to life!