Bricks Detection Dataset

This dataset is intended for brick detection and is organized into two main parts: OBB and Segmentation, corresponding to two different annotation approaches. The OBB part is designed for object detection using oriented bounding boxes. In this format, each image is associated with a text file in which every line represents one annotated object, stored in the standard format as the object class followed by the coordinates of its four corner points. The Segmentation part is designed for pixel-level analysis, where the original bricks are represented in the segmentation masks by gradient ellipses. This part is further divided into three subsets reflecting different levels of scene interpretation. The first subset contains masks in which all bricks, partially visible bricks that cannot be removed by a robot, and the pallet are represented. The second subset contains only the bricks themselves. The third subset contains only the bricks located in the top layer, i.e. the bricks that are accessible and suitable for robotic picking. Within these subsets, the files are split into train, val, and test folders. In the OBB branch, each of these folders contains images and labels subfolders, while in the Segmentation branch, each split contains images and seg subfolders. This structure makes the dataset suitable for both oriented object detection and segmentation-based approaches, while also supporting specialized tasks such as identifying only robotically removable bricks.