FoodLens: Fine-grained and multi-label classification ofIndian Food Images

India has rich cultural diversity which reflects in the variety of food. More than 100k photos are submitted each day onGoogle Maps on Indian restaurants and many more on social media channels. Computer vision has played a key role in parsing food images for automated tagging, nutrition profiling and many more. How-ever, existing state of art food AI-based classification models trained on global food datasets have suboptimal performance on Indian food images. These are attributed to unique challenges in Indian food like lack of representation in food datasets, multiple dishes in the images and fine grained variety of dishes. To address these challenges, we curated an annotated dataset of 30K food images consisting of140+ popular dishes from restaurant menus across India. These images have multi label and fine grained annotations for each dish in the image. All the aggregated popular foods are mapped on a hierarchical tree which models a categorical breakdown of Indian food. We developed a custom deep learning classification model to learn from hierarchy and multilabel information in the Indian food images. CHAMP method shows 5-10% better classification accuracy on Indian food dataset at fine and coarse grained levels compared to state of art food classification models with comparable results for other food bench-mark datasets.