Final Project for IA896 — Introduction to Multimedia Information Retrieval 2016-2 class at FEEC/UNICAMP.
Recognition of the food category and ingredient list of food images.
####Docker
You can find here the docker images (CPU and GPU) used for the project.
####Recognition of Food Categories
The Convolutional Network Inception v3 is used for the recognition of food categories. Starting from the pre-trained model with ImageNet, we perform the fine-tuning of the top layers of this pre-trained network with our dataset of recipes, based on the tutorials:
- How to Retrain Inception's Final Layer for New Categories (code)
- Using TensorBoard to Visualize Image Classification Retraining in TensorFlow
####Ingredient Recognition The Convolutional Network VGG16 is used for our ingredient recognition model. Also starting from the pre-trained model with ImageNet, we perform the fine-tuning based on the tutorial:
- Building powerful image classification models using very little data (Sections 2 and 3)
The pre-trained weights were obtained from this repository. The pre-processing steps for the VGG16, e.g., mean subtraction, load of images, were obtained from keras.
####References