search

dyelax / Adversarial_Video_Generation

A TensorFlow Implementation of "Deep Multi-Scale Video Prediction Beyond Mean Square Error" by Mathieu, Couprie & LeCun.
MIT License
734 stars 184 forks source link

converted code to Tensorflow 1 and changed skimage range #27

Open martinosorb opened 6 years ago

martinosorb commented 6 years ago

The code was converted using automatic conversion provided by Tensorflow.

Additionally, the most recent skimage was telling me its images needed to be between -1 and 1, the opposite of what your code and comments said. I changed that, adding a clip since some values were (very slightly) larger than 1.0. This is slightly suspicious...

Note: I also have a python 3 version, which is NOT included in this PR, but you may be interested in.

dyelax commented 6 years ago

Thanks for the PR! Honestly I haven't touched this repo in a while so I'll need to re-download the data, but I'll try to test this and merge this weekend

dyelax commented 6 years ago

Would also definitely be interested in your Python 3 version!

martinosorb commented 6 years ago

Great, however if I push the py3, it may not be compatible with py2 anymore. Is that ok?

dyelax commented 6 years ago

What did you change to make it work with python3? I’d love to have it be cross-compatible if possible

martinosorb commented 6 years ago

I used 2to3, which does it automatically, and then ran it to check it worked (I had to make one more change before it did). The 2to3 docs state it's usually not a good idea to try making code compatible with both 2 and 3, but that's up to you.