CReLU (Concatenated Rectified Linear Units) with Caffe

CReLU (Understanding and Improving Convolutional Neural Networks via Concatenated Rectified Linear Units)

By Wenling Shang, Kihyuk Sohn, Diogo Almeida, Honglak Lee

Paper : https://arxiv.org/pdf/1603.05201v2.pdf

If you use these models in your research, please cite:

@article{shang2016understanding,
  title={Understanding and Improving Convolutional Neural Networks via Concatenated Rectified Linear Units},
  author={Shang, Wenling and Sohn, Kihyuk and Almeida, Diogo and Lee, Honglak},
  journal={arXiv preprint arXiv:1603.05201},
  year={2016}
}

Results

1-crop validation error on ImageNet (center 224x224 crop from resized image with shorter side=256):

model	top-1	top-5
AlexNet	42.8%	19.7%
SqueezeNet	42.5%	19.7%
CReLU (all) in paper	40.93%	19.39%
CReLU (conv1,4,7) in paper	40.45 %	18.58%
CReLU (conv1–4) in paper	39.82%	18.28%

Donate

(Bitcoin Address) : 3DermWyouJQ5MwoNvEtv2SgnvFRMTDBudv