Add ffmpeg support for unix systems, refactor

Author: gordicaleksa

neural_style_transfer.py

@@ -70,6 +70,7 @@ def neural_style_transfer(config):
         # feature maps need to be of same size for content image and init image
         style_img_resized = utils.prepare_img(style_img_path, np.asarray(content_img.shape[2:]), device)
         init_img = style_img_resized
+
     # we are tuning optimizing_img's pixels! (that's why requires_grad=True)
     optimizing_img = Variable(init_img, requires_grad=True)
 
@@ -80,9 +81,7 @@ def neural_style_transfer(config):
     style_img_set_of_feature_maps = neural_net(style_img)
 
     target_content_representation = content_img_set_of_feature_maps[content_feature_maps_index_name[0]].squeeze(axis=0)
-
     target_style_representation = [utils.gram_matrix(x) for cnt, x in enumerate(style_img_set_of_feature_maps) if cnt in style_feature_maps_indices_names[0]]
-
     target_representations = [target_content_representation, target_style_representation]
 
     # magic numbers in general are a big no no - some things in this code are left like this by design to avoid clutter

utils/utils.py

@@ -41,7 +41,7 @@ def prepare_img(img_path, target_shape, device):
     img = load_image(img_path, target_shape=target_shape)
 
     # normalize using ImageNet's mean
-    # [0, 255] range works much better than [0, 1] range
+    # [0, 255] range worked much better for me than [0, 1] range (even though PyTorch models were trained on latter)
     transform = transforms.Compose([
         transforms.ToTensor(),
         transforms.Lambda(lambda x: x.mul(255)),

utils/video_utils.py

@@ -12,8 +12,8 @@ def create_video_from_intermediate_results(results_path, img_format):
     first_frame = 0
     number_of_frames_to_process = len(os.listdir(results_path))  # default don't trim take process every frame
 
-    ffmpeg = 'ffmpeg.exe'
-    if shutil.which(ffmpeg):  # if ffmpeg.exe is in system path
+    ffmpeg = 'ffmpeg'
+    if shutil.which(ffmpeg):  # if ffmpeg is in system path
         img_name_format = '%' + str(img_format[0]) + 'd' + img_format[1]  # example: '%4d.png' for (4, '.png')
         pattern = os.path.join(results_path, img_name_format)
         out_video_path = os.path.join(results_path, out_file_name)