Abstract:

The primary goal of this project is to divide images into various regions or parts, frequently based on the properties of the pixels in the image. Deep learning systems are more accurate than traditional techniques. Mask R-CNN is used to derive high-level properties from data that are important for machine learning-based semantic segmentation of images. The computer vision method of image segmentation is crucial. To make image analysis simpler, it entails breaking a visual input into segments. Segments are collections of pixels, or "superpixels," that depict objects or portions of objects. There has recently been a significant amount of work targeted at creating image segmentation approaches using deep learning models due to the success of these models in a variety of vision applications. When using CNN to segment images, portions of the picture are fed into the network, and the convolutional neural network labels the pixels as it processes the input. Fully convolutional networks (FCNS) process different input sizes quicker and use convolutional layers to do so. It entails reducing the input image's dimensions before recovering it using orientation invariance skills. The decoder most notable is the R-CNN or region-based convolutional neural networks, and the most recent method called mask R-CNN, which is capable of getting state-of-the-art results on a variety of object detection tasks.

Description:

Semantic segmentation, Instance segmentation, Convolutional neural networks, Deep learning, and Image segmentation