2016/02/03 by Nanne van Noord, Eric Postma, van Noord, Nanne +1
Computer Science · #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning and Data Classification
paper · pdf · doi:10.48550/arxiv.1602.01255
openalex publication_date 2016/02/03 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
Convolutional Neural Networks (CNNs) require large image corpora to be\ntrained on classification tasks. The variation in image resolutions, sizes of\nobjects and patterns depicted, and image scales, hampers CNN training and\nperformance, because the task-relevant information varies over spatial scales.\nPrevious work attempting to deal with such scale variations focused on\nencouraging scale-invariant CNN representations. However, scale-invariant\nrepresentations are incomplete representations of images, because images\ncontain scale-variant information as well. This paper addresses the combined\ndevelopment of scale-invariant and scale-variant representations. We propose a\nmulti- scale CNN method to encourage the recognition of both types of features\nand evaluate it on a challenging image classification task involving\ntask-relevant characteristics at multiple scales. The results show that our\nmulti-scale CNN outperforms single-scale CNN. This leads to the conclusion that\nencouraging the combined development of a scale-invariant and scale-variant\nrepresentation in CNNs is beneficial to image recognition performance.\n