Using HOPE model to train CNNs Without data augmentation: 7.06% error rate on the cifar-10 validation set and 29.47% error rate on the cifar-100 validation set (Single-HOPE-Block)
With data augmentation (rotation+translation+scale+cololr casting): 5.89% error rate on the cifar-10 validation set and 26.99% error rate on the cifar-100 validation set (Single-HOPE-Block).
The paper has been accepted by ACML 2017.
If you hope to use this code, please cite:
@article{pan2016learning,
title={Learning Convolutional Neural Networks using Hybrid Orthogonal Projection and Estimation},
author={Pan, Hengyue and Jiang, Hui},
journal={arXiv preprint arXiv:1606.05929},
year={2016}
}
Based on MatConvNet
MatConvNet is a MATLAB toolbox implementing Convolutional Neural Networks (CNNs) for computer vision applications. It is simple, efficient, and can run and learn state-of-the-art CNNs. Several example CNNs are included to classify and encode images. Please visit the homepage to know more.