Visualize intermediate images of Composer (#244)

* Create a folder for attack.composer.

* Add composer modules for unbounded patch adversary.

* Add config of Adam optimizer.

* Add LoadCoords for patch adversary.

* Add a config of unbounded patch adversary.

* Add a datamodule config for carla patch adversary.

* Fix the simple Linf projection.

* Add composer module PertImageBase for Lp bounded patch adversary.

* Add config of lp-bounded patch adversary.

* Add a fake renderer composer module.

* Teardown a test dataset gracefully for the rendering-in-loop adversary.

* Add configs of simulation-in-loop adversary.

* Add a datamodule config for CARLA patch rendering.

* Update CarlaDataset config.

* Add a composer.visualize switch to see intermediate images.

* Revert "Teardown a test dataset gracefully for the rendering-in-loop adversary."

This reverts commit a5ffef3.

* Revert "Add a composer.visualize switch to see intermediate images."

This reverts commit a17e224.

* Add a composer.visualize switch to see intermediate images.

Author: mzweilin

mart/attack/composer/modular.py

@@ -9,6 +9,7 @@
 from typing import TYPE_CHECKING, Any, Iterable
 
 import torch
+from torchvision.transforms.functional import to_pil_image
 
 from mart.nn import SequentialDict
 
@@ -19,7 +20,7 @@
 
 
 class Composer(torch.nn.Module):
-    def __init__(self, perturber: Perturber, modules, sequence) -> None:
+    def __init__(self, perturber: Perturber, modules, sequence, visualize: bool = False) -> None:
         """_summary_
 
         Args:
@@ -34,6 +35,7 @@ def __init__(self, perturber: Perturber, modules, sequence) -> None:
         if isinstance(sequence, dict):
             sequence = [sequence[key] for key in sorted(sequence)]
         self.functions = SequentialDict(modules, {"composer": sequence})
+        self.visualize = visualize
 
     def configure_perturbation(self, input: torch.Tensor | Iterable[torch.Tensor]):
         return self.perturber.configure_perturbation(input)
@@ -76,6 +78,12 @@ def _compose(
             input=input, target=target, perturbation=perturbation, step="composer"
         )
 
+        # Visualize intermediate images.
+        if self.visualize:
+            for key, value in output.items():
+                if isinstance(value, torch.Tensor):
+                    to_pil_image(value / 255).save(f"{key}.png")
+
         # SequentialDict returns a dictionary DotDict,
         #  but we only need the return value of the most recently executed module.
         last_added_key = next(reversed(output))