Your data team has decided to move outside of the consultation business and into creating its own app! After months of brainstorming and market research you finally identify a gap in the market. You pitch the idea for the app "SeeFood". It's Shazam for food. You get funded immediately. Time to get to work!
- A prototype Streamlit app where a user can upload a picture. Your app will display whether a Hot Dog is present or not.
This link will direct you to your data.
- Start with a basic network architecture, you can add image augmentation and transfer learning as time allows.
- Save your best model to a file and load it into your Streamlit app.
- You might want to have someone focus on building the Streamlit app relatively early in the process.
- In your presentation, tell us your model's improvement over baseline.
- TensorFlow's ImageDataGenerator class can help you load your data.
- Make sure you use a GPU for any CNNs! Kaggle has GPUs available (and maybe TPUs - even faster, but might require code modification). You can choose the upgraded hardware under Settings -> Accelerator. You may need to register and confirm some information first. 🙂
Created by: Greg (Chuck) Dye and adjusted by Jeff Hale