Skip to content

Home

Jump to bottom
Matthew Girard edited this page Mar 24, 2020 · 9 revisions

Welcome to the InverseHalftone wiki!

IMPORTANT KNOWN ISSUES DURING SETUP This algorithm requires a very specific environment and is in the process of being updated to allow more system flexibility. Below are a list of known issues during setup.

  1. Setup of this algorithm will require one traceback to fix. Spyder 3 installation is advisable to address this traceback. The flag allow_pickle must be set to true, but is located in a difficult to navigate to directory. After running main.py for the first time Spyder 3 will provide easy and immediate access to the file by providing a traceback link to it.

  2. VGG19.npy cannot be uploaded to github as the file size exceeds the maximum storage capacity for my github account. The inverse halftone algorithm cannot run without this file. VGG19.npy is a common starting model that can be googled for and found at https://github.com/tensorlayer/pretrained-models/blob/master/models/vgg19.npy, but I can also send via e-mail if this link dies. If you cannot locate then send me a message requesting the file at matthew.girard@uky.edu, and I will send it to you.

  3. Conda install of libraries (numpy, scipy, tensorflow, etc) is first done, then general install of libraries is done (numpy, scipy, tensorflow, etc) in order to work properly with Spyder 3. Best results have been achieved by doing a conda install and pip install in terminal.

  4. Directories must be set in main.py to the corresponding target and raw directories where the normal (target) images are and halftoned (raw) images are contained before running.

  5. Tensorflow.yml file must be run to setup tensorflow (conda) environment before running, otherwise an error will be generated.

  6. The hyperparameters MUST be set and doublechecked before running or else an error will be generated or the deep learning algorithm will produce weird artifacts during verification.

  7. Please use Spyder 3, this setup is tested on Spyder 3.

  8. Keep in mind the InverseHalftone algorithm was written before tensorflow 2, and the yamel file was created using tensorflow 1 so as times go by the versions of certain libraries within the file tensorflow.yml located in this repository might need to be updated if conda returns an error that it cannot locate them. (TODO: Update code to work in newest libraries)

Click here to start.

Clone this wiki locally