Significance and Limitations of Deep Neural Networks for Image Classification and Object Detection

Abstract

Nowadays, Deep Neural Networks are very popular for solving computer vision problems. The motivation of this study is to explore the different deep learning-based architectures used for image classification and object detection. There are various Convolutional Neural Networks(CNNs) based models that perform well for image classification where images are very similar to the dataset. But these models have certain Limitations if tilted or rotated images are processed for classification. Moreover, CNNs are unable to identify the location of different objects in an image. Therefore, to identify the location and recognize the object in the image along with bounding box information, region proposal based object detection algorithms (i.e., R-CNN, Fast R-CNN, Faster R-CNN, and Mask R-CNN) come into the picture. In these object detection algorithms, a bounding box is drawn around the object of interest to locate it within the image. This research article has explored these region-based models along with their pros and cons. Moreover, the performance and purpose of utilization for these object detection-based deep neural network has also been analyzed. Eventually, a summary is provided about these deep neural network-based architectures, where the suitability of a model(s) to solve a particular problem has been discussed.