Merge pull request #27 from Schrodinger-Hat/feat/25

feat(25): make models optionals

ImageGoNord/GoNord.py

@@ -12,11 +12,15 @@
 import ffmpeg
 import uuid
 import shutil
+import requests
 
-import torch
-import skimage.io as io
-import skimage.color as convertor
-import torchvision.transforms as transforms
+try:
+    import torch
+    import skimage.color as convertor
+    import torchvision.transforms as transforms
+except ImportError:
+    # AI feature disabled
+    pass
 
 
 try:
@@ -31,7 +35,12 @@
 from ImageGoNord.utility.quantize import quantize_to_palette
 import ImageGoNord.utility.palette_loader as pl
 from ImageGoNord.utility.ConvertUtility import ConvertUtility
-from ImageGoNord.utility.model import FeatureEncoder,RecoloringDecoder
+
+try:
+    from ImageGoNord.utility.model import FeatureEncoder,RecoloringDecoder
+except ImportError:
+    # AI feature disabled
+    pass
 
 
 class NordPaletteFile:
@@ -158,6 +167,8 @@ class GoNord(object):
     TRANSPARENCY_TOLERANCE = 190
     MAX_THREADS = 10
 
+    PALETTE_NET_REPO_FOLDER = 'https://github.com/Schrodinger-Hat/ImageGoNord-pip/raw/master/ImageGoNord/models/PaletteNet/'
+
     AVAILABLE_PALETTE = []
     PALETTE_DATA = {}
 
@@ -425,6 +436,16 @@ def converted_loop(self, is_rgba, pixels, original_pixels, maxRow, maxCol, minRo
                 pixels[row, col] = tuple(colors_list)
         return pixels
 
+    def load_and_save_models(self):
+        rd_model = requests.get(self.PALETTE_NET_REPO_FOLDER + 'RD.state_dict.pt')
+        fe_model = requests.get(self.PALETTE_NET_REPO_FOLDER + 'FE.state_dict.pt')
+
+        with open(os.path.dirname(palette_net.__file__) + '/FE.state_dict.pt', "wb") as f:
+            f.write(fe_model.content)
+
+        with open(os.path.dirname(palette_net.__file__) + '/RD.state_dict.pt', "wb") as f:
+            f.write(rd_model.content)
+
     def convert_image_by_model(self, image, use_model_cpu=False):
         """
         Process a Pillow image by using a PyTorch model "PaletteNet" for recoloring the image
@@ -444,8 +465,14 @@ def convert_image_by_model(self, image, use_model_cpu=False):
         FE = FeatureEncoder() # torch.Size([64, 3, 3, 3])
         RD = RecoloringDecoder() # torch.Size([530, 256, 3, 3])
 
-        FE.load_state_dict(torch.load(pkg_resources.open_binary(palette_net, "FE.state_dict.pt")))
-        RD.load_state_dict(torch.load(pkg_resources.open_binary(palette_net, "RD.state_dict.pt")))
+        if (
+            os.path.exists(os.path.dirname(palette_net.__file__) + '/FE.state_dict.pt')
+            and os.path.exists(os.path.dirname(palette_net.__file__) + '/RD.state_dict.pt')
+        ):
+            FE.load_state_dict(torch.load(pkg_resources.open_binary(palette_net, "FE.state_dict.pt")))
+            RD.load_state_dict(torch.load(pkg_resources.open_binary(palette_net, "RD.state_dict.pt")))
+        else:
+            self.load_and_save_models()
 
         if use_model_cpu:
             FE.to("cpu")
@@ -472,7 +499,8 @@ def convert_image_by_model(self, image, use_model_cpu=False):
         try:
             pal_np = np.array(palette).reshape(1,6,3)/255
         except:
-            print("You have too many colors in your palette for the model, this feature is limited to 6 colours, now you have: ", len(palette), "! I'll take the first 6!")
+            # this feature is limited to 6 colours
+            # we're taking the first six
             pal_np = np.array(palette[0:6]).reshape(1,6,3)/255
 
         pal = torch.Tensor((convertor.rgb2lab(pal_np) - [50,0,0] ) / [50,128,128]).unsqueeze(0)
@@ -518,7 +546,10 @@ def convert_image(self, image, save_path='', use_model=False, use_model_cpu=Fals
         is_rgba = (image.mode == 'RGBA')
 
         if use_model:
-            image = self.convert_image_by_model(image, use_model_cpu)
+            if torch != None:
+                image = self.convert_image_by_model(image, use_model_cpu)
+            else:
+                print('Please install the dependencies required for the AI feature: pip install image-go-nord[AI]')
         else:
             if not parallel_threading:
                 self.converted_loop(is_rgba, pixels, original_pixels, image.size[0], image.size[1])

ImageGoNord/__init__.py

@@ -1,4 +1,4 @@
 # gonord version
-__version__ = "1.0.2"
+__version__ = "1.1.0"
 
 from ImageGoNord.GoNord import *

index.py

@@ -22,12 +22,10 @@
 # go_nord.add_file_to_palette(NordPaletteFile.AURORA)
 # go_nord.add_file_to_palette(NordPaletteFile.FROST)
 
-# image = go_nord.open_image("images/valley.jpg")
-# go_nord.convert_image(image, save_path="images/test-valley-ai.jpg", use_model=True)
-
-output_path = go_nord.convert_video('videos/SampleVideo_720x480.mp4', 'custom_palette', save_path='videos/SampleVideo_converted.mp4')
-print(output_path)
+image = go_nord.open_image("images/valley.jpg")
+go_nord.convert_image(image, save_path="images/test-valley-ai.jpg", use_model=True)
 exit()
+# output_path = go_nord.convert_video('videos/SampleVideo_720x480.mp4', 'custom_palette', save_path='videos/SampleVideo_converted.mp4')
 
 image = go_nord.open_image("images/test.jpg")
 resized_img = go_nord.resize_image(image)

setup.py

@@ -6,7 +6,7 @@
 
 setup(
     name="image-go-nord",
-    version="1.0.2",
+    version="1.1.0",
     description="A tool to convert any RGB image or video to any theme or color palette input by the user",
     long_description=README,
     long_description_content_type="text/markdown",
@@ -17,7 +17,7 @@
     author_email="schrodinger.hat.show@gmail.com",
     license="AGPL-3.0",
     classifiers=[
-        'Development Status :: 5 - Production/Stable',      # Chose either "3 - Alpha", "4 - Beta" or "5 - Production/Stable" as the current state of your package
+        'Development Status :: 5 - Production/Stable',
         'Intended Audience :: Developers',
         'Topic :: Software Development :: Build Tools',
         "License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)",
@@ -30,8 +30,11 @@
         "Bug Reports": "https://github.com/Schrodinger-Hat/ImageGoNord-pip/issues",
     },
     packages=find_packages(),
-    package_data={'': ['*.txt', 'palettes/*.txt', 'models/*.pt', '*.pt', '*.state_dict.*']},
+    package_data={'': ['*.txt', 'palettes/*.txt']},
     include_package_data=True,
-    install_requires=["Pillow", "ffmpeg-python", "numpy", "torch", "scikit-image", "torchvision"],
+    install_requires=["Pillow", "ffmpeg-python", "numpy", "requests"],
+    extras_require = {
+        'AI':  ["torch", "scikit-image", "torchvision"]
+    },
     python_requires=">=3.5"
 )