template architectures

ganslate/utils/cli/cookiecutter_templates/new_project/{{ cookiecutter.project_name }}/README.md

@@ -0,0 +1 @@
+# {{ cookiecutter.project_name }}

ganslate/utils/cli/cookiecutter_templates/new_project/{{ cookiecutter.project_name }}/architectures/template_architecture.py

@@ -4,7 +4,43 @@
 https://github.com/Maastro-CDS-Imaging-Group/ganslate/tree/master/ganslate/nn
 and to follow the documentation:
 https://ganslate.netlify.app/tutorials/custom_architecture/  # TODO: update the link
+"""
+
+
+# ------------------- Custom GAN from scratch -----------------------
+"""Implementing a custom GAN from scratch is not trivial, and we advise you to go
+through ganslate's GAN source code for an example.
+(https://github.com/Maastro-CDS-Imaging-Group/ganslate/tree/master/ganslate/nn/gans)
+"""
 
- 
+# ------------------ Extending an existing GAN ----------------------
+"""Extending an existing GAN is much easier. This is an example of how you would start
+extending CycleGAN. 
 """
+from ganslate.nn.gans.unpaired import cyclegan
+
+
+@dataclass
+class CustomCycleGANConfig(cyclegan.CycleGANConfig):
+    name: str = "CustomCycleGAN"
+
+
+class CustomCycleGAN(cyclegan.CycleGAN):
+
+    def __init__(self, conf):
+        super().__init__(conf)
+
+    # Now, extend or redefine method(s).
+    # In this example, we extend only the `init_criterions()` method.
+    def init_criterions(self):
+        # Standard GAN loss [Same as in the original CycleGAN]
+        self.criterion_adv = AdversarialLoss(
+            self.conf.train.gan.optimizer.adversarial_loss_type).to(self.device)
+
+        # Fancy loss for generator [Different from the original CycleGAN]
+        self.criterion_G = YourFancyLoss(self.conf)
+
 
+# ------------------ Custom generator or loss -----------------------
+"""No limitations, just basic PyTorch code. They do need to have their corresponding configs
+as can be seen in the documentation and the source code."""