Update processor __call__ for single image/category

processor.py

@@ -76,12 +76,15 @@ class CondViTProcessor(ImageProcessingMixin):
         """
         use_cats = categories is not None
 
-        # Single Image + Single category
+        # Single Image (+ Single category)
         if isinstance(images, Image.Image):
-            images = [images]
+            data = {}
+            data["pixel_values"] = self.process_img(images)
             if use_cats:
-                categories = [categories]
+                data["category_indices"] = self.process_cat(categories)
+            return BatchFeature(data=data)
 
+        # Multiple Images (+ Multiple Categories)
         data = {}
         data["pixel_values"] = torch.stack([self.process_img(img) for img in images])
 
@@ -91,3 +94,4 @@ class CondViTProcessor(ImageProcessingMixin):
             )
 
         return BatchFeature(data=data)
+