Electronic Devices Dataset for 2-Class Semantic Segmentation

This dataset contains images used in the monograph titled Computer Vision in Python. Practical Applications of Deep Learning (Computer vision w Pythonie. Praktyczne zastosowania uczenia głębokiego) to build the U-Net model. The full collection consists of 350 image files of resolution 512x512 pixels showing small electronic devices and office accessories. The dataset was created as follows: after acquiring 100 images, they were randomly split into training (50% of the full dataset), validation (25%), and test (25%) subsets. The training subset was then augmented using vertical and horizontal flips, random rotations in the range of −45° to 45°, and a combination of both flips and random rotation. As a result, the training set contained 300 images, while the validation and test sets each contained 25 images, yielding a total of 350 images. The images are labeled with masks representing 2 kind of objects – REMOTES and BATTERIES. Therefore, the dataset can be used to build models for multiclass semantic segmentation.