Update app.py

app.py

@@ -48,28 +48,28 @@ def display_image(image, size=None, mode='nearest', unnorm=False, title=''):
     
 
 def inferece(num, seed):
-    model_type = 'landscape' 
-    num_im =  int(num)
-    random_seed =  int(seed)
-
+    model_type = 'landscape' #@param ['church', 'face', 'landscape']
+    num_im =  5#@param {type:"number"}
+    random_seed =  90#@param {type:"number"}
+    
     plt.rcParams['figure.dpi'] = 300
-
+    
     mean_latent = load_model(generator, f'{model_type}.pt')
-
+    
     # pad determines how much of an image is involve in the blending
     pad = 512//4
-
+    
     all_im = []
-
+    
     random_state = np.random.RandomState(random_seed)
-
+    
     # latent smoothing
     with torch.no_grad():
         z = random_state.randn(num_im, 512).astype(np.float32)
         z = scipy.ndimage.gaussian_filter(z, [.7, 0], mode='wrap')
         z /= np.sqrt(np.mean(np.square(z)))
-        z = torch.from_numpy(z).cuda()
-
+        z = torch.from_numpy(z).to(device)
+    
         source = generator.get_latent(z, truncation=truncation, mean_latent=mean_latent)
             
         # merge images 2 at a time
@@ -77,25 +77,25 @@ def inferece(num, seed):
             source1 = index_layers(source, i)
             source2 = index_layers(source, i+1)
             all_im.append(generator.merge_extension(source1, source2))
-
+    
     # display intermediate generations    
     # for i in all_im:
     #     display_image(i)
         
-
+    
     b,c,h,w = all_im[0].shape
     panorama_im = torch.zeros(b,c,h,512+(num_im-2)*256)
-
+    
     # We created a series of 2-blended images which we can overlay to form a large panorama
     # add first image
     coord = 256+pad
     panorama_im[..., :coord] = all_im[0][..., :coord]
-
+    
     for im in all_im[1:]:
         panorama_im[..., coord:coord+512-2*pad] = im[..., pad:-pad]
         coord += 512-2*pad
     panorama_im[..., coord:] = all_im[-1][..., 512-pad:]
-
+    
     img = display_image(panorama_im)
     return img