Layout Prediction in Real-world Construction Site Images

The Layout Prediction dataset contains labels for 229 images of real-world construction sites, which are part of the Sequence 2 of the ConSLAM dataset ( https://github.com/mac137/ConSLAM ). This dataset adheres to the conventions defined in SRW-Net ( https://github.com/DavidGillsjo/SRW-Net ).<br>
The dataset includes the original images accompanied by layout annotations. These annotations consist of lines representing various architectural elements, such as walls, ceilings, and doors, with each line specified by coordinates and categorized in a dictionary according to its type. Junctions are also annotated and classified as either "proper" or "false." A "proper" junction indicates an actual endpoint of a line in the real world, while a "false" junction appears to end in the image but does not correspond to a physical endpoint.<br>
The annotated elements are color-coded and stored in a JSON file. The resulting layouts are illustrated in figures, and the Python code used to generate the annotation files is also provided. Please refer to the "Legend.png" file in the dataset for the corresponding color codes of the lables.