MatOR: Material and Object Recognition Dataset

The MatOR dataset comprises 8,354 finely annotated single-object images. Each image is annotated with the object's bounding box, its category, and its material type.The dataset is split into training and test sets in a 7:3 ratio to support model development and evaluation. It covers 14 object categories (e.g., bottle, cup, shirt) and 6 material types, forming 18 real-world, commonly encountered object-material combinations (e.g., plastic bottle, glass cup, ceramic bowl, wooden box). This combination design ensures that the task of material recognition is not a trivial corollary of object category recognition, thereby increasing the challenge and practical significance of the task.In summary, with its carefully designed category-material pairings, realistic data distribution, and controlled object sizes, the MatOR dataset provides a challenging benchmark for research in joint object detection and material recognition.