foodnotfood

A deep learning based web app to classify food and non food images

Live on - https://foodnotfood.herokuapp.com/
Training - https://github.com/vedantthapa/foodnotfood/

Dataset

The data was collected by extracting images from the ImageNet and Food101 datasets.

• Food Images
  • Random subset of Food101
• Non-Food Images
  • 50 Images per class for 1000 classes.
  • Since the classes were randomly downloaded. Some of those classes contained food labels which were then manually moved to the appropriate class.
• Final distribution of the target:
  • Food Images ~ 66%
  • Non-Food Images ~ 34%

References

• https://github.com/mf1024/ImageNet-Datasets-Downloader
• https://github.com/mrdbourke/food-not-food