Fixing Pytorch 2 issues.

Addressing a few bits of code that were causing issues with newer
versions of Pytorch.

Signed-off-by: Matthew Johnson <[email protected]>

.gitignore

@@ -140,4 +140,5 @@ models
 *.ps1
 *.tsv
 figures/*.png
-figures/*.npz
+figures/*.npz
+scenepic.min.js

README.md

@@ -25,9 +25,8 @@ the requirements for the repository, ideally in a virtual environment. We
 recommend using a version of Python >= 3.7. As this code heavily relies upon
 PyTorch, you should install the correct version for your platform. The guide
 [here](https://pytorch.org/get-started/locally/)
-is very useful and I suggest you follow it closely. You may also find
-[this site](https://www.lfd.uci.edu/~gohlke/pythonlibs/) helpful if you 
-are working on Windows. Once that is done, you can run the following:
+is very useful and I suggest you follow it closely. Once that is done,
+you can run the following:
 
 ```
 pip install wheel

fourier_feature_nets/image_dataset.py

@@ -250,11 +250,12 @@ def render(self, samples: RaySamples) -> RenderResult:
         Returns:
             RenderResult: the ground truth render
         """
-        color = self.colors[samples.rays]
+        rays = samples.rays.to(self.colors.device)
+        color = self.colors[rays]
         if self.alphas is None or self.mode == RayDataset.Mode.Dilate:
             alpha = None
         else:
-            alpha = self.alphas[samples.rays]
+            alpha = self.alphas[rays]
             color = torch.where(alpha.unsqueeze(1) > 0, color,
                                 torch.zeros_like(color))
 

fourier_feature_nets/ray_caster.py

@@ -33,7 +33,7 @@
                                    ("train_psnr", float), ("val_psnr", float)])
 
 
-class Raycaster:
+class Raycaster(nn.Module):
     """Implementation of a volumetric raycaster."""
 
     def __init__(self, model: nn.Module):