0.1.5

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

@@ -23,6 +23,8 @@ from ema_pytorch import EMA
 
 from accelerate import Accelerator
 
+from denoising_diffusion_pytorch.version import __version__
+
 # constants
 
 ModelPrediction =  namedtuple('ModelPrediction', ['pred_noise', 'pred_x_start'])
@@ -86,7 +88,10 @@ 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):
     """
@@ -805,8 +810,8 @@ class Trainer(object):
         if self.accelerator.is_main_process:
             self.ema = EMA(diffusion_model, beta = ema_decay, update_every = ema_update_every)
 
-            self.results_folder = Path(results_folder)
-            self.results_folder.mkdir(exist_ok = True)
+        self.results_folder = Path(results_folder)
+        self.results_folder.mkdir(exist_ok = True)
 
         # step counter state
 
@@ -825,7 +830,8 @@ class Trainer(object):
             'model': self.accelerator.get_state_dict(self.model),
             'opt': self.opt.state_dict(),
             'ema': self.ema.state_dict(),
-            'scaler': self.accelerator.scaler.state_dict() if exists(self.accelerator.scaler) else None
+            'scaler': self.accelerator.scaler.state_dict() if exists(self.accelerator.scaler) else None,
+            'version': __version__
         }
 
         torch.save(data, str(self.results_folder / f'model-{milestone}.pt'))
@@ -843,6 +849,9 @@ class Trainer(object):
         self.opt.load_state_dict(data['opt'])
         self.ema.load_state_dict(data['ema'])
 
+        if 'version' in data:
+            print(f"loading from version {data['version']}")
+
         if exists(self.accelerator.scaler) and exists(data['scaler']):
             self.accelerator.scaler.load_state_dict(data['scaler'])
 

denoising_diffusion_pytorch/version.py

@@ -0,0 +1 @@
+__version__ = '0.1.5'

setup.py

@@ -1,9 +1,11 @@
 from setuptools import setup, find_packages
 
+exec(open('denoising_diffusion_pytorch/version.py').read())
+
 setup(
   name = 'denoising-diffusion-pytorch',
   packages = find_packages(),
-  version = '0.31.1',
+  version = __version__,
   license='MIT',
   description = 'Denoising Diffusion Probabilistic Models - Pytorch',
   author = 'Phil Wang',