Web-based application that uses deep learning models directly in the browser to identify food from an image (classifier), estimate its volume (segmentation), and provide a rough calorie count. This project demonstrates the power of running AI tasks entirely on the client side using TensorFlow.js.
URL Webapp
Still in early stages (needs training on larger epochs, better calorie count, better food db, better/newer arch) but functional.
- Frontend = HTML, CSS, JavaScript, Tailwind
- Machine Learning Frameworks = TensorFlow, Keras, Tensorflow.js
- Notebooks used to train and convert the model
- Food101 - Classification -
Food101 - Food103 - Segmentation -
EduardoPacheco/FoodSeg103
Vast AI was used to train the models. The command is
vastai create instance <OFFER_ID> --image vastai/tensorflow:2.16.1-cuda-12.4.1 --env '-p 1111:1111 -p 6006:6006 -p 8080:8080 -p 8384:8384 -p 72299:72299 -e OPEN_BUTTON_PORT=1111 -e OPEN_BUTTON_TOKEN=1 -e JUPYTER_DIR=/ -e DATA_DIRECTORY=/workspace/ -e PORTAL_CONFIG="localhost:1111:11111:/:Instance Portal|localhost:8080:18080:/:Jupyter|localhost:8080:8080:/terminals/1:Jupyter Terminal|localhost:8384:18384:/:Syncthing|localhost:6006:16006:/:Tensorboard"' --onstart-cmd 'entrypoint.sh' --disk 200 --jupyter --ssh --direct
- Train higher epochs
- Optimizing food db