🎲 Support for setting the random seed.

Author: alexjc

src/texturize/__main__.py

@@ -6,6 +6,7 @@
  |_| |_|\___|\__,_|_|  \__,_|_|  \__\___/_/\_\\__|\__,_|_|  |_/___\___|
 
 Usage:
+    texturize SOURCE... [--size=WxH] [--output=FILE] [--seed=SEED] [--device=DEVICE]
                         [--octaves=O] [--precision=P] [--iterations=I]
     texturize --help
 
@@ -18,6 +19,7 @@
 Options:
     SOURCE                  Path to source image to use as texture.
     -s WxH, --size=WxH      Output resolution as WIDTHxHEIGHT. [default: 640x480]
+    --seed=SEED             Configure the random number generation.
     --device=DEVICE         Hardware to use, either "cpu" or "cuda".
     --octaves=O             Number of octaves to process. [default: 5]
     --precision=P           Set the quality for the optimization. [default: 1e-4]
@@ -96,7 +98,7 @@ def _prepare_gram(self, features):
 
 
 def get_all_layers(critics):
-    """Determine the minimal list of layer features that needs to be extracted from the image.
+    """Determine the minimal list of features that needs to be extracted from the image.
     """
     layers = set(itertools.chain.from_iterable(c.get_layers() for c in critics))
     return sorted(list(layers))
@@ -323,6 +325,13 @@ def main():
     # Scan all the files based on the patterns specified.
     files = itertools.chain.from_iterable(glob.glob(s) for s in config["SOURCE"])
     for filename in files:
+        # If there's a random seed, use it for all images.
+        if config["--seed"] is not None:
+            seed = int(config["--seed"])
+            torch.manual_seed(seed)
+            torch.cuda.manual_seed(seed)
+
+        # By default, disable autograd until the core optimization loop.
         with torch.no_grad():
             try:
                 run(config, filename)