Novel Framework for Real-Time Semantic Image Segmentation

Abstract

Today Computer Vision has taken a major turn in the Artificial Intelligence domain. The image segmentation technique, which is frequently based on the attributes of the image’s pixels, is the most extensively used approach in computer vision for dividing an image into multiple portions or regions. In this paper, we present a thorough examination of our semantic segmentation model developed for the classroom scenario. We created a dataset with over 200 class objects, such as chairs, tables, whiteboards, books, pens, and other classroom items, and trained our model on it to segment classroom images accurately. To accurately segment images and achieve a high level of accuracy, our model employs cutting-edge deep learning techniques like the convolutional neural networks (CNNs) and attention mechanisms. The model obtained an overall accuracy of 90% on the test set, indicating its ability to appropriately segment and identify items in a classroom scenario. Overall, our semantic segmentation model’s results on the 200 classes of classroom environment dataset show that it has the potential to improve safety, accessibility, and organization in educational settings.