Segmentation For Object-Based Image Analysis (OBIA) Using Tensorflow Framework

Abstract

Deep learning is the ultimate breakthrough of artificial intelligence and it will change the world dramatically in this century. Various type of deep neural networks has been used to resolve challenging computer vision problems such as detection, localization, recognition and segmentation of objects in the wild. Semantic segmentation to separate a portrait from the video background. Semantical Segmentation. This process essentially closes the image bits based on an object class to vacuum them together. In this paper we differentiate four separate deep learning models that we trained to provide real - time webcam video segmentation of portrait images, and analyze performance on the respective phases. They employ two distinct deep learning architectures and they employ two different datasets. Their data sets more than 30,000 human portrait images each. ​TensorFlow and Keras – This is how we train our models (approach 2). Architecture 1 takes RGB images of 256×256 as input with around 12–14 FPS at runtime compared to Architecture 2 which takes RGB images of128 ×128 with runtimes at around15-18FPS. In our work, we outlined a principled method that earned greater accuracy and efficiency than any method described. Tags: portrait segmentation, semantic segmentation, TensorFlow in Flutter is the process of integrating TensorFlow, an open-source machine learning framework widely used for building machine learning applications, into a Flutter app. TensorFlow gives powerful library for model building and deployment for machine learning application, while flutter is a library for creating good-looking, well-designed interfaces.