Todo

Next: On branch cleanup cleaning the code for the main app, next is to make the data downloading more robust.

• [ ] Create a new branch to keep cleaning up
• [ ] Data ideas
  • Food images = Food5k + Food101 (random subset)
  • Non-food images = Food5k + Open Images (random subset of 10000 images from training set)
• [ ] Clean up data download
  • [ ] Use Open Images for non-food images:
    • Download images from open images: https://www.kaggle.com/getting-started/157163
    • Downloaded images from open images v2: https://github.com/monocongo/openimages
    • Another dataset for food and non-food images: https://www.kaggle.com/binhminhs10/food5k
    • Can download images from Open Images using FiftyOne — https://voxel51.com/docs/fiftyone/tutorials/open_images.html
  • [ ] Use Open Images + Food101 for food images:
    • Can also filter on Open Image for food images?
      • Download random set of images from Open Images — filter food & not food
      • Use Food101 random subset to create images of Food + could also download images from the internet of bulk food classes to classify food
• [ ] Clean up model training to reflect changes on data
  • [ ] Make model training scriptable
• [ ] Clean up GitHub repo to make the app reproducible
  • [ ] README
  • [ ] Turn notebooks into scripts
• [ ] Fix prediction probablity for foodnotfood.app

Rules

• Every hour:
  • 10 pushups, 10 squats, 10 kicks, 10 punches
  • Surprises at 77-minutes in

Hours

• [x] 1
• [x] 2
• [x] 3
• [x] 4
• [x] 5
• [x] 6
• [x] 7
• [x] 8
• [x] 9