add options

app.py

@@ -8,19 +8,25 @@ path = "./smoot.mp4"
 interpolator = Interpolator()
 
 
-def predict(image_a, image_b):
+def predict(image_a, image_b, preview):
     image1 = load_image(image_a)
     image2 = load_image(image_b)
     input_frames = [image1, image2]
-    frames = list(interpolate_recursively(input_frames, interpolator))
-    mediapy.write_video(path, frames, fps=30)
+    if preview:
+        fps = 3
+        frames = interpolator.preview_frames(input_frames)
+    else:
+        fps = 30
+        frames = list(interpolate_recursively(input_frames, interpolator))
+
+    mediapy.write_video(path, frames, fps=fps)
     return path
 
 
 footer = r"""
 <center>
 <b>
-Demo for <a href='https://www.tensorflow.org/hub/tutorials/tf_hub_delf_module'>DELF</a>
+Demo for <a href='https://www.tensorflow.org/hub/tutorials/tf_hub_film_example'>FILM model</a>
 </b>
 </center>
 """
@@ -33,20 +39,26 @@ coffee&emoji=&slug=leonelhs&button_colour=FFDD00&font_colour=000000&font_family=
 </center>
 """
 
-with gr.Blocks(title="DELF") as app:
-    gr.HTML("<center><h1>Match images using DELF</h1></center>")
-    gr.HTML("<center><h3>Neural network and logic for processing images to identify keypoints and their "
-            "descriptors.</h3></center>")
+with gr.Blocks(title="FILM") as app:
+    gr.HTML("<center><h1>Frame interpolation using the FILM model</h1></center>")
+    gr.HTML("<center><h3>Frame interpolation is the task of synthesizing many in-between images from a given set of "
+            "images. The technique is often used for frame rate upsampling or creating slow-motion video "
+            "effects.</h3></center>")
     with gr.Row(equal_height=False):
         with gr.Column():
-            input_img_a = gr.Image(type="filepath", label="Input image A")
-            input_img_b = gr.Image(type="filepath", label="Input image B")
+            with gr.Row(equal_height=True):
+                with gr.Column():
+                    input_img_a = gr.Image(type="filepath", label="Input image A")
+                with gr.Column():
+                    input_img_b = gr.Image(type="filepath", label="Input image B")
+            pre = gr.Checkbox(label="Preview", value=True, info="Run in preview mode video")
             run_btn = gr.Button(variant="primary")
+
         with gr.Column():
-            output_img = gr.Video(format="mp4", label="Interpolate video")
+            output_img = gr.Video(format="mp4", label="Interpolate video", autoplay=True)
             gr.ClearButton(components=[input_img_a, input_img_b, output_img], variant="stop")
 
-    run_btn.click(predict, [input_img_a, input_img_b], [output_img])
+    run_btn.click(predict, [input_img_a, input_img_b, pre], [output_img])
 
     with gr.Row():
         blobs_a = [[f"examples/image_a/{x:02d}.jpg"] for x in range(1, 2)]

interpolator.py

@@ -1,7 +1,8 @@
+from typing import Generator, List, Iterable
+
 import numpy as np
 import tensorflow as tf
-import tensorflow_hub as hub
-from typing import Generator, List, Iterable
+from huggingface_hub import snapshot_download
 
 """A wrapper class for running a frame interpolation based on the FILM model on TFHub
 
@@ -12,6 +13,8 @@ Usage:
   (B,H,W,C) layout, batch_dt is the sub-frame time in range [0..1], (B,) layout.
 """
 
+FILM_REPO_ID = "leonelhs/film"
+
 
 def _pad_to_align(x, align):
     """Pads image batch x so width and height divide by align.
@@ -63,7 +66,9 @@ class Interpolator:
         inference.'
     """
         self.times_to_interpolate = times_to_interpolate
-        self._model = hub.load("https://tfhub.dev/google/film/1")
+        model_path = snapshot_download(FILM_REPO_ID)
+        self._model = tf.saved_model.load(model_path)
+        # self._model = hub.load("https://tfhub.dev/google/film/1")
         self._align = align
 
     def __call__(self, x0: np.ndarray, x1: np.ndarray,
@@ -92,6 +97,18 @@ class Interpolator:
             image = tf.image.crop_to_bounding_box(image, **bbox_to_crop)
         return image.numpy()
 
+    def preview_frames(self, frames: List[np.ndarray]):
+
+        time = np.array([0.5], dtype=np.float32)
+
+        media_input = {
+            'time': np.expand_dims(time, axis=0),  # adding the batch dimension to the time
+            'x0': np.expand_dims(frames[0], axis=0),  # adding the batch dimension to the image
+            'x1': np.expand_dims(frames[1], axis=0)  # adding the batch dimension to the image
+        }
+        mid = self._model(media_input)
+        return [frames[0], mid['image'][0].numpy(), frames[1]]
+
 
 def _recursive_generator(
         frame1: np.ndarray, frame2: np.ndarray, num_recursions: int,

requirements.txt

@@ -1,3 +1,4 @@
+gradio
 tensorflow>=2.15.0
 tensorflow-hub>=0.15.0
 requests>=2.31.0