version 0.0.1

Author: jun

README.md

@@ -3,13 +3,13 @@
 Simple torch module implementation of [Alias-Free GAN](https://nvlabs.github.io/alias-free-gan/).
 
 This repository including
-- Alias-Free GAN style lowpass sinc filter @[filter.py](/filter.py)
+- Alias-Free GAN style lowpass sinc filter @[filter.py](/src/alias_free_torch/filter.py)
 
-- Alias-Free GAN style up/downsample @[resample.py](/resample.py)
+- Alias-Free GAN style up/downsample @[resample.py](/src/alias_free_torch/resample.py)
 
-- Alias-Free activation @[act.py](/act.py)
+- Alias-Free activation @[act.py](/src/alias_free_torch/act.py)
 
-- and test codes @[./test](/test)
+- and test codes @[./test](/src/alias_free_torch/test)
 
 **Note: Since this repository is unofficial, filter and upsample could be different with [official implementation](https://github.com/NVlabs/stylegan3).**
 

pyproject.toml

@@ -0,0 +1,6 @@
+[build-system]
+requires = [
+    "setuptools>=42",
+    "wheel"
+]
+build-backend = "setuptools.build_meta"

setup.py

@@ -0,0 +1,34 @@
+from setuptools import setup, find_packages
+
+with open("README.md", "r", encoding="utf-8") as fh:
+    long_description = fh.read()
+
+
+setup(
+    name                = 'alias_free_torch',
+    version             = '0.0.1',
+    description         = 'alias free torch',
+    long_description=long_description,
+    long_description_content_type="text/markdown",
+    author              = 'junjun3518',
+    author_email        = 'junjun3518@gmail.com',
+    url                 = 'https://github.com/junjun3518/alias-free-torch',
+    install_requires    =  [],
+    packages            = find_packages(where = "src"),
+    package_dir         = {"": "src"},
+    keywords            = ['alias','torch','pytorch','filter'],
+    python_requires     = '>=3',
+    zip_safe            = False,
+    classifiers         = [
+        'Programming Language :: Python :: 3',
+        'Programming Language :: Python :: 3.2',
+        'Programming Language :: Python :: 3.3',
+        'Programming Language :: Python :: 3.4',
+        'Programming Language :: Python :: 3.5',
+        'Programming Language :: Python :: 3.6',
+        'Programming Language :: Python :: 3.7',
+        'Programming Language :: Python :: 3.8',
+        'Programming Language :: Python :: 3.9',
+    ],
+)
+

src/__init__.py

@@ -0,0 +1,2 @@
+
+

src/alias_free_torch/__init__.py

@@ -0,0 +1,3 @@
+from .filter import LowPassFilter1d, LowPassFilter2d
+from .resample import UpSample1d, UpSample2d
+from .act import Activation1d, Activation2d

act.py src/alias_free_torch/act.py

@@ -6,8 +6,8 @@
 if 'sinc' in dir(torch):
     sinc = torch.sinc
 else:
-# This code is adopted from adefossez's julius.core.sinc
-# https://adefossez.github.io/julius/julius/core.html
+    # This code is adopted from adefossez's julius.core.sinc
+    # https://adefossez.github.io/julius/julius/core.html
     def sinc(x: torch.Tensor):
         """
         Implementation of sinc, i.e. sin(pi * x) / (pi * x)
@@ -26,8 +26,9 @@ def __init__(self,
                  half_width=0.6,
                  stride: int = 1,
                  pad: bool = True,
-                 kernel_size=12):  # kernel_size should be even number for stylegan3 setup,
-                                   # in this implementation, odd number is also possible.
+                 kernel_size=12
+                 ):  # kernel_size should be even number for stylegan3 setup,
+        # in this implementation, odd number is also possible.
         super().__init__()
         if cutoff < -0.:
             raise ValueError("Minimum cutoff must be larger than zero.")
@@ -54,7 +55,7 @@ def __init__(self,
         if self.even:
             time = (torch.arange(-self.half_size, self.half_size) + 0.5)
         else:
-            time = torch.arange(self.kernel_size) - self.half_size 
+            time = torch.arange(self.kernel_size) - self.half_size
         if cutoff == 0:
             filter_ = torch.zeros_like(time)
         else:
@@ -71,11 +72,10 @@ def forward(self, x):
         new_shape = shape[:-1] + [-1]
         x = x.view(-1, 1, shape[-1])
         if self.pad:
-            x = F.pad(
-                x,
-                (self.half_size, self.half_size),
-                mode='constant', value=0) # empirically, it is better than replicate
-                #mode='replicate')
+            x = F.pad(x, (self.half_size, self.half_size),
+                      mode='constant',
+                      value=0)  # empirically, it is better than replicate
+            #mode='replicate')
         if self.even:
             out = F.conv1d(x, self.filter, stride=self.stride)[..., :-1]
         else:
@@ -90,7 +90,7 @@ def __init__(self,
                  stride: int = 1,
                  pad: bool = True,
                  kernel_size=12):  # kernel_size should be even number
-                                   # in this implementation, odd number is also possible.
+        # in this implementation, odd number is also possible.
         super().__init__()
         if cutoff < -0.:
             raise ValueError("Minimum cutoff must be larger than zero.")
@@ -115,15 +115,15 @@ def __init__(self,
 
         #rotation equivariant grid
         if self.even:
-        time = (torch.stack(torch.meshgrid(
-            torch.arange(-self.half_size, self.half_size) + 0.5,
-            torch.arange(-self.half_size, self.half_size) + 0.5),
-                            dim=-1))
+            time = torch.stack(torch.meshgrid(
+                torch.arange(-self.half_size, self.half_size) + 0.5,
+                torch.arange(-self.half_size, self.half_size) + 0.5),
+                               dim=-1)
         else:
-        time = (torch.stack(torch.meshgrid(
-            torch.arange(self.kernel_size) - self.half_size,
-            torch.arange(self.kernel_size) - self.half_size,
-                            dim=-1))
+            time = torch.stack(torch.meshgrid(
+                torch.arange(self.kernel_size) - self.half_size,
+                torch.arange(self.kernel_size) - self.half_size),
+                               dim=-1)
 
         time = torch.norm(time, dim=-1)
         #rotation equivariant window
@@ -148,10 +148,12 @@ def forward(self, x):
         shape = list(x.shape)
         x = x.view(-1, 1, shape[-2], shape[-1])
         if self.pad:
-            x = F.pad(x, (self.half_size, self.half_size, self.half_size,
-                          self.half_size),
-                      mode='constant', value=0) # empirically, it is better than replicate or reflect
-                      #mode='replicate')
+            x = F.pad(
+                x, (self.half_size, self.half_size, self.half_size,
+                    self.half_size),
+                mode='constant',
+                value=0)  # empirically, it is better than replicate or reflect
+            #mode='replicate')
         if self.even:
             out = F.conv2d(x, self.filter, stride=self.stride)[..., :-1, :-1]
         else:

filter.py src/alias_free_torch/filter.py

@@ -1,8 +1,8 @@
 import torch
 import matplotlib.pyplot as plt
-from ..act import Activation2d
-from ..resample import UpSample2d, DownSample2d
-from ..filter import LowPassFilter2d
+from alias_free_torch.act import Activation2d
+from alias_free_torch.resample import UpSample2d, DownSample2d
+from alias_free_torch.filter import LowPassFilter2d
 import math
 continuous_ratio = 16
 ratio = 2

resample.py src/alias_free_torch/resample.py

@@ -1,6 +1,6 @@
 import torch
 import matplotlib.pyplot as plt
-from ..resample import DownSample1d
+from alias_free_torch.resample import DownSample1d
 
 ratio = 10
 t = torch.arange(100) / 100. * 3.141592

test/act2d_test.py

@@ -1,6 +1,6 @@
 import torch
 import matplotlib.pyplot as plt
-from ..resample import DownSample2d
+from alias_free_torch.resample import DownSample2d
 
 ratio = 4
 size = 80

test/down1d_test.py

@@ -1,6 +1,6 @@
 import torch
 import matplotlib.pyplot as plt
-from ..filter import LowPassFilter1d
+from alias_free_torch.filter import LowPassFilter1d
 
 ratio = 2
 t = torch.arange(400) / 40. * 3.141592

test/down2d_test.py

@@ -1,6 +1,6 @@
 import torch
 import matplotlib.pyplot as plt
-from ..filter import LowPassFilter2d
+from alias_free_torch.filter import LowPassFilter2d
 
 ratio = 2
 

test/filter1d_test.py

@@ -1,7 +1,7 @@
 import torch
 import matplotlib.pyplot as plt
-from ..resample import UpSample1d
-from ..filter import LowPassFilter1d
+from alias_free_torch.resample import UpSample1d
+from alias_free_torch.filter import LowPassFilter1d
 
 ratio = 2
 t = torch.arange(100) / 10. * 3.141592

test/filter2d_test.py

@@ -1,7 +1,7 @@
 import torch
 import matplotlib.pyplot as plt
-from ..resample import UpSample2d
-from ..filter import LowPassFilter2d
+from alias_free_torch.resample import UpSample2d
+from alias_free_torch.filter import LowPassFilter2d
 
 ratio = 8
 size = 40