fix data not being normalized to range of -1 to 1

denoising_diffusion_pytorch/denoising_diffusion_pytorch.py

@@ -190,6 +190,8 @@ class Unet(nn.Module):
         channels = 3
     ):
         super().__init__()
+        self.channels = channels
+
         dims = [channels, *map(lambda m: dim * m, dim_mults)]
         in_out = list(zip(dims[:-1], dims[1:]))
 
@@ -229,7 +231,7 @@ class Unet(nn.Module):
                 Upsample(dim_in) if not is_last else nn.Identity()
             ]))
 
-        out_dim = default(out_dim, 3)
+        out_dim = default(out_dim, channels)
         self.final_conv = nn.Sequential(
             Block(dim, dim),
             nn.Conv2d(dim, out_dim, 1)
@@ -291,11 +293,13 @@ class GaussianDiffusion(nn.Module):
         denoise_fn,
         *,
         image_size,
+        channels = 3,
         timesteps = 1000,
         loss_type = 'l1',
         betas = None
     ):
         super().__init__()
+        self.channels = channels
         self.image_size = image_size
         self.denoise_fn = denoise_fn
 
@@ -389,7 +393,8 @@ class GaussianDiffusion(nn.Module):
     @torch.no_grad()
     def sample(self, batch_size = 16):
         image_size = self.image_size
-        return self.p_sample_loop((batch_size, 3, image_size, image_size))
+        channels = self.channels
+        return self.p_sample_loop((batch_size, channels, image_size, image_size))
 
     @torch.no_grad()
     def interpolate(self, x1, x2, t = None, lam = 0.5):
@@ -450,7 +455,8 @@ class Dataset(data.Dataset):
             transforms.Resize(image_size),
             transforms.RandomHorizontalFlip(),
             transforms.CenterCrop(image_size),
-            transforms.ToTensor()
+            transforms.ToTensor(),
+            transforms.Lambda(lambda t: (t * 2) - 1)
         ])
 
     def __len__(self):
@@ -548,7 +554,8 @@ class Trainer(object):
                 batches = num_to_groups(36, self.batch_size)
                 all_images_list = list(map(lambda n: self.ema_model.sample(batch_size=n), batches))
                 all_images = torch.cat(all_images_list, dim=0)
-                utils.save_image(all_images, str(RESULTS_FOLDER / f'sample-{milestone}.png'), nrow=6)
+                all_images = (all_images * 0.5) + 1
+                utils.save_image(all_images, str(RESULTS_FOLDER / f'sample-{milestone}.png'), nrow = 6)
                 self.save(milestone)
 
             self.step += 1

setup.py

@@ -3,7 +3,7 @@ from setuptools import setup, find_packages
 setup(
   name = 'denoising-diffusion-pytorch',
   packages = find_packages(),
-  version = '0.6.0',
+  version = '0.6.4',
   license='MIT',
   description = 'Denoising Diffusion Probabilistic Models - Pytorch',
   author = 'Phil Wang',