Food Hunting with CNNs

This script does two things:

• Outputs the probability of the input image containing food
• Localizes the food in the image

Usage

python findfood.py INPUT_IMAGE_FILENAME OUTPUT_IMAGE_FILENAME

ex. python findfood.py ./bread.jpg ./breadout.jpg

You can download the model from here: [https://s3-us-west-2.amazonaws.com/models-a-c-ozbek/model_food.h5] Put the model file in the same level folder of the script.

Brief Technical Explanation

• By using Food-5K dataset (http://mmspg.epfl.ch/food-image-datasets), a food vs not-food model is trained by fine-tuning InceptionV3 in Keras
• [http://cnnlocalization.csail.mit.edu] The technique here is implemented for food localization. (Class Activation Mapping)
• You may check out the notebooks folder for more detail.

Results