Hand-drawn Power Converter Circuit Diagrams

Citation Author(s):
Bharat
Bohara
University of Houston
Harish S.
Krishnamoorthy
University of Houston
Submitted by:
Bharat Bohara
Last updated:
Wed, 01/24/2024 - 17:26
DOI:
10.21227/e2tg-zj02
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

The dataset is created on June 8, 2023, at 12:28 AM, consists of 760 annotated images in Pascal VOC format. The dataset was exported via roboflow.com, an end-to-end computer vision platform that facilitates various tasks such as collaboration, image collection and organization, data understanding and search, annotation, dataset creation, model training and deployment, and active learning for dataset improvement. Each image in the dataset underwent pre-processing, which involved auto-orienting the pixel data by removing EXIF orientation, and resizing the images to a uniform size of 640x640 pixels through stretching. To increase the dataset's diversity and variability, the following augmentations were applied to generate five versions of each source image: horizontal flip (50% probability), vertical flip (50% probability), 90-degree rotations (none, clockwise, counter-clockwise, upside-down with equal probability), random rotation (-15 to +15 degrees), random shear (-15° to +15° horizontally and vertically), random brightness adjustment (-25% to +25%), random exposure adjustment (-25% to +25%), random Gaussian blur (0 to 1.25 pixels), and the application of salt and pepper noise to 3% of the pixels. Researchers and developers can utilize this dataset and access state-of-the-art computer vision training.