A deep learning system that classifies food images into 230 categories and retrieves matching recipes using convolutional neural networks.
Food-Recipe-CNN is a deep learning system that uses convolutional neural networks to classify food images into 230 categories and retrieve matching recipes. It solves the problem of automated food recognition where visually similar dishes from different categories overlap, by combining CNN-based classification with nearest-neighbor search. The system processes over 400,000 food images and 300,000 recipes to provide accurate recipe suggestions.
Developers and researchers working on computer vision, deep learning, or food recognition projects, as well as those interested in building applications that link images to textual data like recipes.
It offers a novel method that combines CNN classification with nearest-neighbor search to improve recipe retrieval accuracy, using the largest German-language recipe dataset available and dynamically determined categories via semantic analysis.
food image to recipe with deep convolutional neural networks.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Uses over 300,000 recipes from chefkoch.de, providing a comprehensive foundation for training and retrieval, as mentioned in the dataset description.
Combines CNN classification with approximate nearest neighbor search to improve accuracy on visually similar dishes, addressing overlaps in food categories as described in the solution process.
Employs topic modeling and semantic analysis to dynamically determine 230 food categories, enhancing classification relevance based on recipe names.
Leverages transfer learning with models like AlexNet, VGG, ResNet, and GoogLeNet, saving training time and resources while utilizing state-of-the-art networks.
Code comments and the recipe dataset are primarily in German, limiting accessibility and usability for developers who don't speak the language.
The DeepChef web application is explicitly noted as 'work in progress,' meaning it's not fully functional or ready for deployment without additional development effort.
Requires processing over 400,000 images and running complex CNNs with PCA, necessitating significant GPU resources and technical expertise, which can be a barrier for smaller teams.