Initial version 🥳

app.py

@@ -9,20 +9,19 @@ from skimage.transform import resize
 from skimage import img_as_ubyte
 from skimage.color import rgb2gray
 
-
-from tensorflow import keras
+from huggingface_hub.keras_mixin import from_pretrained_keras
 
 # load model
-model = keras.models.load_model('saved_model')
+model = from_pretrained_keras("keras-io/conv-lstm")
 
 # Examples
 samples = []
-example_driving = os.listdir('asset/source')
-for video in example_driving:
+example_source = os.listdir('asset/source')
+for video in example_source:
     samples.append([f'asset/source/{video}', 0.5, True])
 
 
-def inference(driving,
+def inference(source,
               split_pred = 0.4, # predict 0.6% of video
               predict_one = False, # Whether to predict a sliding one frame or all frames at once
               output_name = 'output.mp4',
@@ -30,19 +29,19 @@ def inference(driving,
               cpu = False,
               ):
 
-    # driving
-    reader = imageio.get_reader(driving)
+    # source
+    reader = imageio.get_reader(source)
     fps = reader.get_meta_data()['fps']
-    driving_video = []
+    source_video = []
     try:
         for im in reader:
-            driving_video.append(im)
+            source_video.append(im)
     except RuntimeError:
         pass
     reader.close()
-    driving_video = [rgb2gray(resize(frame, (64, 64)))[..., np.newaxis] for frame in driving_video]
+    source_video = [rgb2gray(resize(frame, (64, 64)))[..., np.newaxis] for frame in source_video]
     
-    example = np.array(driving_video)
+    example = np.array(source_video)
     print(example.shape)
     # Pick the first/last ten frames from the example.
     start_pred_id = int(split_pred * example.shape[0]) # prediction starts from frame start_pred_id