Intersection over Union based analysis of Image detection/segmentation using CNN model

Abstract

Neural networks are capable of learning high-dimensional hierarchical structures of objects from huge quantities Deep-learning systems can learn to recognize photographs based on a large amount of training data. Artificial intelligence has this as one of its features. Deep-learning algorithms for picture interpretation may be divided into two groups. SegNet, U-Net, and SharpMask are examples of fully convolutional methods that use an encoder-decoder architecture. Region-based methods, on the other hand, use a convolutional neural network (CNNs) stack to extract features, such as Mask-RCNN, PSP Net and DeepLab. When the networks are trained on a large enough number of annotated datasets, region-based methods beat for most image segmentation tasks, fully convolutional techniques are used. We designed and incorporated deep-learning techniques based on Mask-RCNN to detect 2D images while creating a segmentation for each mask item in this paper.