Name: GalaxyGAN
Owner: SpaceML
Description: null
Created: 2016-11-29 11:44:02.0
Updated: 2017-08-12 02:45:58.0
Pushed: 2017-07-25 16:29:09.0
Homepage: null
Size: 32091
Language: JavaScript
GitHub Committers
| User | Most Recent Commit | # Commits |
|---|---|---|
| Hantian Zhang | 2017-07-25 12:43:38.0 | 1 |
| Gokula Krishnan | 2017-01-07 17:12:23.0 | 3 |
| Seth Ariel Green | 2017-07-06 15:17:54.0 | 1 |
| Yiru Chen | 2017-02-21 14:50:28.0 | 4 |
Other Committers
| User | Most Recent Commit | # Commits | |
|---|---|---|---|
| Hantian Zhang | hantian@public-docking-cx-0414.ethz.ch | 2017-02-03 14:52:39.0 | 1 |
| Hantian Zhang | hantian@public-docking-cx-0568.ethz.ch | 2017-02-08 15:52:33.0 | 1 |
| Hantian Zhang | hantian@public-docking-cx-2860.ethz.ch | 2016-11-30 21:18:18.0 | 2 |
| Hantian Zhang | hantian@public-docking-cx-4041.ethz.ch | 2016-12-05 16:10:32.0 | 13 |
| Hantian Zhang | hantian@public-docking-hpx-3692.ethz.ch | 2016-11-29 12:05:31.0 | 5 |
| Hantian Zhang | hantian@sgd-hanzhang-01.ethz.ch | 2017-07-25 16:29:05.0 | 6 |
| Ubuntu | ubuntu@ip-172-31-19-171.ec2.internal | 2016-12-05 15:27:19.0 | 6 |
We provide an EC2 AMI with the following pre-installed packages:
as well as the FITS file we used in the paper(saved in ~/fits_train and ~/fits_test)
AMI Id: ami-6f48b379 . (Can be launched using p2.xlarge instance in GPU compute catagory)
Launch an instance.
Please follow the instruction of Amazon EC2.
note: If you get error like “nvidia-uvm 4.4.0-62 generic” was missing, this is because Amazon updated the kernal of the Ubuntu system, please re-install the cuda again.
license.lic in ~/ of the EC2 instance you createdbash /usr/local/MATLAB/R2016b/bin/activate_matlab.sh -propertiesFile /home/ubuntu/activate.txt to active matlabgit clone --recursive https://github.com/SpaceML/GalaxyGAN.git
Please execute the following three commands and you will get the result that we got in our paper.
cd GalaxyGAN
bash train.sh -input ~/fits_train -fwhm 1.4 -sigma 1.2 -figure figures -gpu 1 -model models
This will run the trainning on all FITS files in ~/fits_train.
bash test.sh -input ~/fits_test -fwhm 1.4 -sigma 1.2 -figure figures -gpu 1 -output result -model models -mode full
This will run the testing on all FITS files in ~/fits_test and you can see the results in result/1.4_1.2/latest_net_G_test/.
bash test.sh -input ~/fits_test -fwhm 1.4 -sigma 1.2 -figure figures -gpu 1 -output result -model models -mode blank
This will run the testing on all FITS files in ~/fits_test, just the groundtruth is made blank to make sure the testing doesn't use the information of groundtruth image that we provide. In the figures/test/ you can see the groundtruth are left blank and you can still get the same output in result/1.4_1.2/latest_net_G_test/ as the previous command.
You can vary the parameters after -fwhm and -sigma to change the variance of gaussian filter and white noise level to the number you want.
It will take about 5 hours to train the model on an Amazon EC2 p2.xlarge instance.
Linux or OSX
NVIDIA GPU + CUDA CuDNN (CPU mode and CUDA without CuDNN may work with minimal modification, but untested)
Install torch and dependencies from https://github.com/torch/distro
Install torch packages nngraph, threads and display
luarocks install nngraph
luarocks install threads
luarocks install https://raw.githubusercontent.com/szym/display/master/display-scm-0.rockspec
Install matlab
Install imagemagick
git clone --recursive https://github.com/SpaceML/GalaxyGAN.git
cd GalaxyGAN
The data to download is about 5GB, after unzipping it will become about 16GB.
bash download.sh
Please execute the following three commands and you will get the result that we got in our paper.
bash train.sh -input fitsdata/fits_train -fwhm 1.4 -sigma 1.2 -figure figures -gpu 1 -model models
This will run the trainning on all FITS files in fitsdata/fits_train.
bash test.sh -input fitsdata/fits_test -fwhm 1.4 -sigma 1.2 -figure figures -gpu 1 -output result -model models -mode full
This will run the testing on all FITS files in fitsdata/fits_test and you can see the results in result/1.4_1.2/latest_net_G_test/.
bash test.sh -input fitsdata/fits_test -fwhm 1.4 -sigma 1.2 -figure figures -gpu 1 -output result -model models -mode blank
This will run the testing on all FITS files in fitsdata/fits_test, just the groundtruth is made blank to make sure the testing doesn't use the information of groundtruth image that we provide. In the figures/test/ you can see the groundtruth are left blank and you can still get the same output in result/1.4_1.2/latest_net_G_test/ as the previous command.
You can vary the parameters after -fwhm and -sigma to change the variance of gaussian filter and white noise level to the number you want.
It will take about 5 hours to train the model on an Amazon EC2 p2.xlarge instance.