Create utils.py

utils.py

@@ -0,0 +1,38 @@
+import tensorflow as tf
+import numpy as np
+from einops import rearrange
+from decord import VideoReader
+
+num_frames = 16
+input_size = 224
+patch_size = (16, 16)
+IMAGENET_MEAN = np.array([0.45, 0.45, 0.45])
+IMAGENET_STD = np.array([0.225, 0.225, 0.225])
+
+def format_frames(frame, output_size):
+    frame = tf.image.convert_image_dtype(frame, tf.uint8)
+    frame = tf.image.resize(frame, size=output_size)
+    frame = frame / 255.
+    frame = frame - IMAGENET_MEAN
+    frame = frame / IMAGENET_STD
+    return frame
+
+def read_video(file_path):
+    container = VideoReader(file_path)
+    return container
+
+def frame_sampling(container, num_frames):
+    interval = len(container) // num_frames
+    bids = np.arange(num_frames) * interval
+    offset = np.random.randint(interval, size=bids.shape)
+    frame_index = bids + offset
+    frames = container.get_batch(frame_index).asnumpy()
+    frames = np.stack(frames)
+    frames = format_frames(frames, [input_size] * 2)
+    return frames
+
+def denormalize(image):
+    image = image.numpy() if not isinstance(image, np.ndarray) else image
+    image = image * IMAGENET_STD + IMAGENET_MEAN
+    image = (image * 255).clip(0, 255).astype('uint8')
+    return image