Side-scan sonar boulder recognition dataset

The Side-scan sonar boulder recognition dataset consists of 3 classes "Boulder", "Shadow", and "Boulder+Shadow" created to use the combined shape for localisation of the entire entity while Boulder and Shadow shapes can be used to create a separate model or algorithm separating boulder and its shadow for further analysis or estimation of dimensions. The resulting dataset consists of 2000 boulder instances located in 1845 images. The data was into training, validation, and test sets at the point of creation for the purpose of inference using sonar swaths which cannot be published. The exact distribution is as follows: Train - 1478 images - 1600 instances (each class) Test - 183 images - 200 instances (each class) Valid - 184 images - 200 instances (each class) All images are saved in greyscale PNG format where each pixel is a single sample representing the intensity of the signal received by the sonar. The sonar data was collected from three undisclosed locations.