A Review of Fixed Input Size Limitation in Convolutional Neural Networks Models and Proposed Solutions

Abstract

Convolutional Neural Networks (CNNs) are incredibly powerful deep learning techniques that have been applied
to computer vision applications to yield innovative results. CNNs are ideal for applications like object
identification, image segmentation, and image classification because they can automatically extract pertinent
information from the images without human supervision. While CNNs can attain state-of-the-art performance in
many applications and domains, most CNNs currently have limitations in training and prediction due to their
sensitivity to image size. As a result, image recognition datasets are typically downsized to the input size
specification of the CNN models. This study's objective is to examine CNN models and suggest possible
solutions to tackle the fixed input size problem that exists in CNN models.