Update app.py

app.py

@@ -85,52 +85,50 @@ def get_tiles(img, mode=0):
             result.append({'img':img3[i], 'idx':i})
         return result, n_tiles_with_info >= n_tiles
 
-def getitem(img,tile_mode):
-        sub_imgs=False
-        
-        tiff_file = img
-        image = skimage.io.MultiImage(tiff_file)[0]
-        tiles, OK = get_tiles(image, tile_mode)
-
-        idxes=n_tiles
-        idxes = np.asarray(idxes)
-        idxes=np.random.choice(list(range(n_tiles)),n_tiles, replace=False)
-
-        n_row_tiles = int(np.sqrt(n_tiles))
-        images = np.zeros((image_size * n_row_tiles, image_size * n_row_tiles, 3))
-        for h in range(n_row_tiles):
-            for w in range(n_row_tiles):
-                i = h * n_row_tiles + w
-    
-                if len(tiles) > idxes[i]:
-                    this_img = tiles[idxes[i]]['img']
-                else:
-                    this_img = np.ones((image_size, image_size, 3)).astype(np.uint8) * 255
-                this_img = 255 - this_img
-                h1 = h * image_size
-                w1 = w * image_size
-                images[h1:h1+image_size, w1:w1+image_size] = this_img
-
-#         images = 255 - images
-        images = images.astype(np.float32)
-        images /= 255
-        images = images.transpose(2, 0, 1)
-
-        return torch.tensor(images)
+def getitem(img, tile_mode):
+    tiff_file = img
+    image = skimage.io.MultiImage(tiff_file)[0]
+    tiles, OK = get_tiles(image, tile_mode)
+
+    idxes = np.random.choice(list(range(n_tiles)), n_tiles, replace=False)
+
+    n_row_tiles = int(np.sqrt(n_tiles))
+    images = np.zeros((image_size * n_row_tiles, image_size * n_row_tiles, 3))
+    for h in range(n_row_tiles):
+        for w in range(n_row_tiles):
+            i = h * n_row_tiles + w
+
+            if len(tiles) > idxes[i]:
+                this_img = tiles[idxes[i]]['img']
+            else:
+                this_img = np.ones((image_size, image_size, 3)).astype(np.uint8) * 255
+            this_img = 255 - this_img
+            h1 = h * image_size
+            w1 = w * image_size
+            images[h1:h1 + image_size, w1:w1 + image_size] = this_img
+
+    images = images.astype(np.float32)
+    images /= 255
+    images = images.transpose(2, 0, 1)
+
+    # Add a batch dimension
+    return torch.tensor(images).unsqueeze(0)
+
+
 def predict_label(im):
-    data1=getitem(im,0)
-    data2=getitem(im,2)
-    LOGITS=[]
-    LOGITS2=[]
+    data1 = getitem(im, 0)
+    data2 = getitem(im, 2)
+    LOGITS = []
+    LOGITS2 = []
+
     with torch.no_grad():
         data1 = data1.to(device)
         logits = models[0](data1)
         LOGITS.append(logits)
 
-
-        data = data.to(device)
-        logits = models[0](data)
-        LOGITS2.append(logits)
+        data2 = data2.to(device)
+        logits2 = models[0](data2)
+        LOGITS2.append(logits2)
 
     LOGITS = (torch.cat(LOGITS).sigmoid().cpu() + torch.cat(LOGITS2).sigmoid().cpu()) / 2
     PREDS = LOGITS.sum(1).round().numpy()
@@ -138,7 +136,6 @@ def predict_label(im):
 
 
 
-
 def classify_images(im):
     pred,idx,probs=predict_label(im)
     s='Your submitted case has Prostate cancer of ISUP Grade '+pred