Researchers in China have created an artificial intelligence model capable of recognizing dishes, their flavors, and nutritional content solely from images. The development involved approximately 2,000 studio images of 32 popular Chinese meals, including chicken, pork, and beef dishes. The team examined the chemical makeup of each meal and utilized this data to train a multimodal deep learning model that associates the visual characteristics of food with its nutritional composition. Remarkably, the AI successfully identified the dish in 91.5% of cases and provided nutritional information regarding protein, fat, and carbohydrates that closely matched laboratory analysis. During testing, volunteers shared their dietary preferences and calorie needs, allowing the algorithm to recommend suitable meals, which received a rating of 8 out of 9 for accuracy. The creators aim to implement this technology in the food industry for quality assurance, using a camera to detect and reject substandard portions before they are packaged. Additionally, they aspire to develop an application that can analyze an image of a plate to estimate calorie content and suggest dishes, although this will necessitate a larger dataset. Would you feel comfortable letting AI choose your lunch?
Informational material. 18+.
