Initialization Using Perlin Noise for Training Networks with a Limited Amount of Data

01/19/2021
by   Nakamasa Inoue, et al.
11

We propose a novel network initialization method using Perlin noise for training image classification networks with a limited amount of data. Our main idea is to initialize the network parameters by solving an artificial noise classification problem, where the aim is to classify Perlin noise samples into their noise categories. Specifically, the proposed method consists of two steps. First, it generates Perlin noise samples with category labels defined based on noise complexity. Second, it solves a classification problem, in which network parameters are optimized to classify the generated noise samples. This method produces a reasonable set of initial weights (filters) for image classification. To the best of our knowledge, this is the first work to initialize networks by solving an artificial optimization problem without using any real-world images. Our experiments show that the proposed method outperforms conventional initialization methods on four image classification datasets.

READ FULL TEXT

page 3

page 4

page 5

research
11/08/2022

Learning advisor networks for noisy image classification

In this paper, we introduced the novel concept of advisor network to add...
research
07/26/2017

Graph-Based Classification of Omnidirectional Images

Omnidirectional cameras are widely used in such areas as robotics and vi...
research
09/15/2017

Joint Hierarchical Category Structure Learning and Large-Scale Image Classification

We investigate the scalable image classification problem with a large nu...
research
08/22/2020

Analyzing the Contribution of Training Algorithms on Deep Neural Networks for Hyperspectral Image Classification

Autoencoder (AE)-based deep neural networks learn complex problems by ge...
research
06/24/2020

Learning Interclass Relations for Image Classification

In standard classification, we typically treat class categories as indep...
research
06/05/2022

A statistical analysis of an image classification problem

The availability of massive image databases resulted in the development ...
research
03/15/2018

Neural Network Quine

Self-replication is a key aspect of biological life that has been largel...

Please sign up or login with your details

Forgot password? Click here to reset