Computing technologies for food analysis have recently attracted significant attention due to their potential to support applications that promote healthier lifestyles, such as diet monitoring and food recommendation systems. These tools can help reduce the risk of diet-related diseases. A key component in developing such applications is the automatic recognition of food images, which alleviates the need for users to manually log consumed or available ingredients. While numerous datasets exist for prepared food, there are few publicly available resources that focus specifically on food ingredients. In this work, we present a dataset comprising approximately 4,000 densely annotated images of food ingredients and food products, intended to support the development of food-related applications. Each image includes instance segmentation annotations, capturing food items as they might appear on refrigerator shelves or tables, thus providing rich contextual information for recognition tasks. We present a detailed analysis of the dataset’s composition. Then, to assess its quality and utility, we report experimental results using several neural network models for instance segmentation.