Skip to content

statsml/Equiv-FCL-CONVL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

47 Commits
 
 
 
 
 
 

Repository files navigation

Equiv-FCL-CONVL

An equivalence of fully connected layer and convolutional layer.

Dependence packages

  * numpy
  * python3
  * tensorflow
  * keras
  * panda
  * h5py
  * matplotlib
  * skimage

Python files

  * trainnetworks.py        # train CNN and FC network.
  * visiualNet.py           # plot the architecture of the networks.
  * computeFnorm.py         # compare the two well tuned networks, plot historams of the weights and filters.
  * net.py                  # define CNN and FC network.
  * img2col.py              # converting 4D data to 2D matrix.
  * Data.py                 # data provider.
  * plotcsv.plotHistory.py  # plot the training and validation loss.
  * logger.BachLosses.py    # record the loss of every batch druing training.

Running programs

  • train the two networks python3 trainnetworks.py. The log file and model are stored in the directory logs and model.
  • visiualize the two networks python3 visualNet.py. The reulsts are stored in the logs directory.
  • compare the two well-tuned networks, python3 computeFnorm.py.
  • visualize the losses of the two networks, python3 plotHistory.py.

Authors

Cite out technical report if it is useful to your research:

@misc{1712.01252,
  author = {Wei Ma and Jun Lu},
  title = {{A}n {E}quivalence of {F}ully {C}onnected {L}ayer and {C}onvolutional {L}ayer},
  year = {2017},
  eprint = {1712.01252}
}

References

  1. Program hipsternet
  2. Andrea Vedaldi and Karel Lenc. Matconvnet: Convolutional neural networks for matlab. In Proceedings of the 23rd ACM international conference on Multimedia, pp. 689–692. ACM, 2015.

Releases

No releases published

Packages

No packages published

Languages