Faster region based convolution neural network with context iterative refinement for object detection
Kishore Anthuvan Sahayaraj K., G. Balamurugan
Abstract
In this paper, proposed a novel method to improve the localisation precision of identified objects. We present a framework for iteratively enhancing image region recommendations to meet ground truth values in this research. The Faster R–CNN (FR-CNN) seems to be an object recognition deep convolutional network. It gives the user the impression that the network is cohesive and single. The network can provide accurate and timely predictions about the whereabouts of a range of objects. We first build a unified model based on rapid predictions to relocate inaccurate area recommendations. Because the emphasis is on object detection, it may be utilized with a wide range of datasets and is compatible with various FR-CNN architectures. Second, we focus on the application of the joint score function to a variety of picture features. This joint score function depicts the location of the concealed object concerning other objects. The picture data and an updated structured production loss function are the only two inputs that influence the parameters of the joint scoring function. The join-score function and iterative context refinement (CIR) are used to generate our final unified model, which is then classified using Smooth Support Vector Machine (SSVM). We measured accuracy using the mean average precision after training FR-CNN + CIR and SSVM on a low-cost GPU using the PASCAL VOC 2012 dataset. Our results are 3.6 % more exact than rival deep learning algorithms on average.