GitXplorerGitXplorer
y

HFM_PyTorch

public
5 stars
4 forks
0 issues

Commits

List of commits on branch main.
Verified
b3964490f4256ef25ae1db278fd7f58edaba304a

Update README.md

yyiranwu0 committed 3 years ago
Verified
621a9ac45803132d3a90a524d8a8ac664f5eea64

Update README.md

yyiranwu0 committed 4 years ago
Verified
910b928f5b39a55c952bcadc81924f9bd21c1945

Add files via upload

yyiranwu0 committed 4 years ago
Verified
bc9d9beee965ff78f30d77b92846c00f984f4b4d

Update README.md

yyiranwu0 committed 4 years ago
Verified
0a35d58fcc3797fae03de5ac8d4ff65e789e6cb5

Update README.md

yyiranwu0 committed 4 years ago
Verified
ff7b518b87b712f017686720985c818094072ce7

Update README.md

yyiranwu0 committed 4 years ago

README

The README file for this repository.

HFM_PyTorch

An alternate version of HFM code with PyTorch. Right now only utilities.py and Cylinder2D_flower_systematic.py are presented.

Additional files

  • DataManager.py: to save training loss and errors during training.
  • test.py: to test the results.
  • plot.py: to plot the results.

Notes:

The results are in Results folder. Comparing to the original code, this error rate of this version is a bit higher.
However, further training can be done by using a smaller learning rate to achieve better error rate.
A learning rate of 1e-3 (used in the original code) is prone to overfitting and the results are bad.
A learning rate of 1e-4 (showed in results)can have similar results comparing to the original paper.

To train:

  • download data and place it into a Data folder.

python Cylinder2D_flower_systematic.py 201 15000 [cuda-device-num|optional] [using visdom|optional]

[cuda-device-num]: don't need if not using GPU.
[using visdom] : this was not tested, but was planned to use visdom to visualize training process.

To test:

  • You should have related files in Results folder.
    python test.py v10 [cuda-device-num|optional]

To plot:

  • You should have related files in Results folder.
    python plot.py v10