SeeFood
SeeFood is a food image classifier trained on image data of hot dogs and other food items to classify whether food in the picture is a hotdog or not.This project was inspired by the seefood App in HBO's Silicon Valley show . To build this I used a pre-trained Densenet121 model and applied transfer learning with the help of GPUs in cloud to fine tune the model to the hot-dogs dataset. This was made using PyTorch with Python. Made into a web application with Flask
Deployed on Heroku : Gradio demo
Model
- Classifier built upon a densenet121 model using transfer learning using PyTorch.
- Dataset , Hot Dog - Not Hot Dog from kaggle
- The model gave a accuracy of 87% unseen test data.
Run locally
Clone the repo
git clone https://github.com/vinayakj02/SEEFOOD-classifier.git
Change the working directory to SEEFOOD-classifier
cd SEEFOOD-classifier
Install the requirements
python3 -m pip install -r requirements.txt
Start the server
python3 app.py
Run with docker
Clone the repo
git clone https://github.com/vinayakj02/SEEFOOD-classifier.git
Change the working directory to SEEFOOD-classifier
cd SEEFOOD-classifier
Build the image
sudo docker build -t seefood .
Run the container
sudo docker run -it -d -p 7000:7000 seefood
Visit localhost:7000 to view the site.