HealthifyMe, a leading Indian health and wellness startup backed by Khosla Ventures, has unveiled a groundbreaking AI-powered feature that allows users to automatically recognize Indian food from images for efficient calorie intake logging. This new functionality, called Snap, revolutionizes the way users track their meals, making the process more convenient and accurate.
Indian health and wellness startup, HealthifyMe, introduces an AI-powered feature called Snap that allows users to automatically recognize Indian food from images for efficient calorie intake logging. This innovative technology simplifies the process of food tracking, increases user engagement, and offers accurate suggestions for food items, ensuring a seamless user experience.
Challenges in Tracking Indian Foods
Tracking Indian foods through images poses unique challenges due to the vast variety of cuisines and the complexity of traditional Indian plates called Thaalis. These plates consist of different portions of various food items, making it difficult for a model to recognize them accurately alongside their portion sizes. However, HealthifyMe’s Snap feature addresses this challenge.
How Snap Works
Snap allows users to capture pictures of their meals and automatically identifies the food items within the images. Alternatively, users can grant access to their photo gallery, and the app will scan food pictures automatically. Later, users can review these pictures and the food items within them, enabling them to log their calorie intake at their convenience.
The model on the device detects the food pictures and sends them to servers for specific dish recognition, ensuring user privacy. HealthifyMe has found that the gallery-based model works more effectively as it has more time to recognize the food items compared to the option of taking pictures of the meal for recognition.
Improved User Experience and Accuracy
If the model detects multiple items in an image, Snap prompts users to tap on each item and add it to their calorie tracking. Users will see an adjustable rectangular box to focus on different items, ensuring accurate tracking.
HealthifyMe’s Co-founder & CEO, Tushar Vashisht, highlighted that the company’s primary use case for its free users has always been food tracking. By introducing Snap, HealthifyMe aims to increase user engagement and retention by simplifying the food tracking process, eliminating the need for users to physically type and remember to log their food.
The company had previously attempted to implement image-based food recognition but faced challenges. However, the availability of advanced generative AI models made developing Snap possible. Currently, the feature is trained to recognize 150,000 Indian food items, with a 60-70% accuracy rate. Even if the model fails to accurately recognize a food item, users receive suggestions about what the item could possibly be, ensuring a seamless experience.
HealthifyMe employs human reviewers who review and correct false recognitions. Additionally, users can manually tag falsely recognized photos to further improve the accuracy of the model. Vashisht expressed confidence that the accuracy will increase to over 80% in the next month.
Competition and Future Plans
HealthifyMe faces competition in the field of AI-powered food recognition, with companies like Samsung’s meal planning platform Samsung Food and Snapcalorie, a startup led by former Google Lens engineer Wade Norris, also working on solving the same problem.
In the coming weeks, HealthifyMe plans to offer users more options for food logging. Users can now send pictures to HealthifyMe’s WhatsApp or tag them on X with a food image. The company is also developing a voice input feature and enhancing its existing AI-powered assistant, Ria.
With its continuous efforts to enhance its technology, HealthifyMe aims to revolutionize the way people track their meals and maintain a healthy lifestyle. As the company partners with food and grocery services, it expects to expand its reach and provide its groundbreaking technology to a larger audience.