Retune ESRGAN hyperparameters on stronger discriminator

[weiji14]

Retuning hyperparameters as there's been some significant changes made since #92. Main push was because I noticed some the discriminator was implemented somewhat differently compared with ESRGAN's. But also, there's been a new dataset added at #112 and a new Tesla V100 GPU (0a1f719) to train on so might as well!

TODO:

• Patch Enhancing the Super Resolution Generative Adversarial Network #78 after a closer review at the code:
  • implementing discriminator to be the same as ESRGAN's (6f8d7ae)
  • Fix bug with content loss and adversarial loss weighing being accidentally swapped (d3a5b74)
• Tune hyperparameters (6f46087)
• Set new hyperparameter defaults (465e95d)

[review-notebook-app]

Check out this pull request on ReviewNB: https://app.reviewnb.com/weiji14/deepbedmap/pull/129

Visit www.reviewnb.com to know how we simplify your Jupyter Notebook workflows.

[weiji14]

Weird problem with rasterio breaking the build, even though we have not changed any software versions?!! Error message as follows:

---------------------------------------------------------------------------
CPLE_OpenFailedError                      Traceback (most recent call last)
rasterio/_base.pyx in rasterio._base.DatasetBase.__init__()

rasterio/_shim.pyx in rasterio._shim.open_dataset()

rasterio/_err.pyx in rasterio._err.exc_wrap_pointer()

CPLE_OpenFailedError: 'highres/2010tr.nc' not recognized as a supported file format.

During handling of the above exception, another exception occurred:

RasterioIOError                           Traceback (most recent call last)
<ipython-input-13-34f5dd97040f> in <module>
      1 hireses = [
      2     selective_tile(filepath=f"highres/{f}", window_bounds=w)
----> 3     for f, w in zip(filepaths, window_bounds)
      4 ]
      5 hires = np.concatenate(hireses)

<ipython-input-13-34f5dd97040f> in <listcomp>(.0)
      1 hireses = [
      2     selective_tile(filepath=f"highres/{f}", window_bounds=w)
----> 3     for f, w in zip(filepaths, window_bounds)
      4 ]
      5 hires = np.concatenate(hireses)

<ipython-input-3-07c3751fe850> in selective_tile(filepath, window_bounds, padding, out_shape, gapfill_raster_filepath)
     32     array_list = []
     33 
---> 34     with rasterio.open(filepath) as dataset:
     35         print(f"Tiling: {filepath}")
     36         for window_bound in window_bounds:

~/.local/share/virtualenvs/deepbedmap-qAKEkYde/lib/python3.6/site-packages/rasterio/env.py in wrapper(*args, **kwds)
    419 
    420         with env_ctor(session=session):
--> 421             return f(*args, **kwds)
    422 
    423     return wrapper

~/.local/share/virtualenvs/deepbedmap-qAKEkYde/lib/python3.6/site-packages/rasterio/__init__.py in open(fp, mode, driver, width, height, count, crs, transform, dtype, nodata, sharing, **kwargs)
    214         # None.
    215         if mode == 'r':
--> 216             s = DatasetReader(path, driver=driver, **kwargs)
    217         elif mode == 'r+':
    218             s = get_writer_for_path(path)(path, mode, driver=driver, **kwargs)

rasterio/_base.pyx in rasterio._base.DatasetBase.__init__()

RasterioIOError: 'highres/2010tr.nc' not recognized as a supported file format.