fixed discriminator

models/discriminator.py

@@ -15,8 +15,8 @@ def conv3d(in_channel:int, out_channel:int):
  nn.Module: Conv3d Module
  """
  return nn.Sequential(
- nn.Conv3d(channel[0], channel[1], kernel_size=4, stride=2, padding=1, bias=False),
- nn.BatchNorm3d(channel[1]),
+ nn.Conv3d(in_channel, out_channel, kernel_size=4, stride=2, padding=1, bias=False),
+ nn.BatchNorm3d(out_channel),
  nn.LeakyReLU(0.2, inplace=True),
  )
 

requirements.txt

@@ -1,4 +1,3 @@
-fancy_einsum
 torch
 torchvision
 tqdm

utils/utils.py

@@ -55,17 +55,17 @@ def preprocess(images, sizex=None, sizey=None):
 
 
 
-def downsample(mat:np.ndarray, down_sample=64)->np.ndarray:
+def downsample(mat:np.ndarray, new_size=64)->np.ndarray:
  """
  Args:
  mat (np.ndarray): 3d image 
- down_sample (int, optional): the new number of voxels cubes. Defaults to 64.
+ new_size (int, optional): the new number of voxels cubes. Defaults to 64.
 
  Returns:
  np.ndarray: 3d images after downsampling
  """
- lost_dim = int(mat.shape[2]/down_sample)
- return mat.reshape((-1, down_sample, lost_dim, down_sample, lost_dim, down_sample, lost_dim)).mean(
+ lost_dim = int(mat.shape[2]/new_size)
+ return mat.reshape((-1, new_size, lost_dim, new_size, lost_dim, new_size, lost_dim)).mean(
  axis=(2, 4, 6)
  )
 
@@ -97,4 +97,4 @@ def load_data(data_path:str, down_sample=None):
  img.append(cv2.imread(folder_path + "/" + format(i, "03d") + ".png")) # pylint: disable=no-member
  img2d.append(preprocess(np.array(img)))
  img3d.append(mat_data)
- return img2d, img3d
+ return img2d, img3d