xueshengke / pyramidCNN

This project is an attempt to ultilize the layer-wise training procedure, i.e. pyramid, to learn a convolutional neural network. However, the performances on accuracy and time are not desirable on different image datasets. Therefore, this project is deprecated for now.

Our proposal of this idea originates from this paper:

@article{
王冠皓徐军-39,
   Author = {王冠皓 and 徐军},
   Title = {基于多级金字塔卷积神经网络的快速特征表示方法},
   Journal = {计算机应用研究},
   Number = {08},
   Pages = {2492-2495},
   Note = {页数: 4},
   Abstract = {由于在大尺度图像中卷积滤波的过程速度过慢,导致卷积神经网络(convolutional neural network,CNN)参数调节困难、训练时间过长。针对这一问题,通过对传统卷积神经网络(traditional convolution neural network,TCNN)的改进,提出一种快速有效的多级金字塔卷积神经网络(multi-level pyramid CNN,MLPCNN)。这一网络使用权值共享的方法将低级的滤波权值共享到高级,保证CNN的训练只在较小尺寸的图像块上进行,加快了训练速度。实验表明,在特征维数比较低的情况下,MLPCNN提取到的特征比传统的特征提取方法更加有效,在...},
   Keywords = {深度学习;多级金字塔卷积神经网络;特征表示;特征共享},
   Year = {2015} }

Perhaps it is not completely believable.

--------------
If you have any questions about this project, please do not hesitate to contact me.
Shengke, Xue
e-mail: [e-mail], [e-mail], or [e-mail]