mawiszus / world-gan Goto Github PK

Hi,

it's not working with Python 3.9, 3.8 cause bpy 2.82.1 (Requires: Python >=3.7, <3.8). Than I could import most of the requirements with Python 3.7.9.

But ...

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
pytorch-lightning 1.2.2 requires PyYAML!=5.4.*,>=5.1, but you have pyyaml 5.4 which is incompatible.

Than I installed

pip install PyYAML==5.3.1

But when I try to run it, I get:

python main.py *TODO* --alpha 100 --niter 4000 --nfc 64

Traceback (most recent call last):
  File "main.py", line 3, in <module>
    from generate_samples import generate_samples
  File "D:\projects\github\Mawiszus\World-GAN\generate_samples.py", line 7, in <module>
    from config import Config
  File "D:\projects\github\Mawiszus\World-GAN\config.py", line 4, in <module>
    from typing import List, Literal, Optional, Union
ImportError: cannot import name 'Literal' from 'typing' (C:\Users\rafal\AppData\Local\Programs\Python\Python37\lib\typing.py)

Which Python version is working for this project?

Thx

Hi,
Can you please explain about the data and the preprocessing steps as well as code for it? I am trying to use my own data.
You response would be very helpful for me.
Thank you

Hi,

I'm currently working on an undergraduate Honors thesis building upon the work that you've done, but have been having some issues and was wondering if you might be able to clarify part of this code. I have found that when training the GAN on my own area, the generated area seems to always be nearly identical to the original. I've done some exploration of the code and found that this line seems to imply that there is no noise being applied at other scales, despite the comment saying that noise should be added to the previous output. Could you please clarify how noise is handled at scales > 0? The fact that this appears to be no noise at other scales would explain why my generated worlds are nearly identical to the original, but I'm not confident in how I can modify this to get much more variation. Thank you for your help!

Excuse me, in the config.py, I set input_dir and input_name to 'input/minecraft' and 'Empty_World' respectively. However, when I run the training, it will have an error message: 'input/minecraft/Empty_World/region/r.2.2.mca'. I've verified that this file doesn't exist in that path. Can you please advise on how to resolve this issue?

Hi again!

I've made a lot of progress, but I'm having an error when trying to train a block2vec embedding using your existing code. I'm getting the error: AttributeError: Can't pickle local object 'create_compund_nbt_class.<locals>.CompundNBTTag', which I've been able to trace as a result of the CompoundNBTTag class being declared locally within PyAnvilEditor.

I was curious how you got around this error. I've tried using the exact same PyTorch and PyTorchLightning versions that you have in the requirements, but there has been no change. I've learned that this error comes from PyTorch Lightning trying to use multiprocessing for the training, but have yet to find a way around the error.

Any help would be appreciated! Thank you!