trust paper

README.md

@@ -42,7 +42,7 @@ diffusion = GaussianDiffusion(
     loss_type = 'l1'    # L1 or L2
 )
 
-training_images = torch.randn(8, 3, 128, 128) # images are normalized from 0 to 1
+training_images = torch.rand(8, 3, 128, 128) # images are normalized from 0 to 1
 loss = diffusion(training_images)
 loss.backward()
 # after a lot of training
@@ -124,7 +124,7 @@ diffusion = GaussianDiffusion1D(
     objective = 'pred_v'
 )
 
-training_seq = torch.randn(8, 32, 128) # features are normalized from 0 to 1
+training_seq = torch.rand(8, 32, 128) # features are normalized from 0 to 1
 loss = diffusion(training_seq)
 loss.backward()
 
@@ -248,3 +248,13 @@ sampled_seq.shape # (4, 32, 128)
     volume  = {abs/2207.12598}
 }
 ```
+
+```bibtex
+@article{Sunkara2022NoMS,
+    title   = {No More Strided Convolutions or Pooling: A New CNN Building Block for Low-Resolution Images and Small Objects},
+    author  = {Raja Sunkara and Tie Luo},
+    journal = {ArXiv},
+    year    = {2022},
+    volume  = {abs/2208.03641}
+}
+```

denoising_diffusion_pytorch/denoising_diffusion_pytorch.py

@@ -86,7 +86,10 @@ def Upsample(dim, dim_out = None):
     )
 
 def Downsample(dim, dim_out = None):
-    return nn.Conv2d(dim, default(dim_out, dim), 4, 2, 1)
+    return nn.Sequential(
+        Rearrange('b c (h p1) (w p2) -> b (c p1 p2) h w', p1 = 2, p2 = 2),
+        nn.Conv2d(dim * 4, default(dim_out, dim), 1)
+    )
 
 class WeightStandardizedConv2d(nn.Conv2d):
     """

setup.py

@@ -3,7 +3,7 @@ from setuptools import setup, find_packages
 setup(
   name = 'denoising-diffusion-pytorch',
   packages = find_packages(),
-  version = '0.31.1',
+  version = '0.32.0',
   license='MIT',
   description = 'Denoising Diffusion Probabilistic Models - Pytorch',
   author = 'Phil Wang',