| import tensorflow as tf |
| import numpy as np |
| from einops import rearrange |
| from decord import VideoReader |
|
|
| num_frames = 32 |
| input_size = 224 |
| patch_size = (16, 16) |
| IMAGENET_MEAN = np.array([123.675, 116.28, 103.53]) |
| IMAGENET_STD = np.array([58.395, 57.12, 57.375]) |
|
|
| 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 - 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(z): |
| mean = np.array([123.675, 116.28, 103.53]) |
| variance = np.array([np.square(58.395), np.square(57.12), np.square(57.375)]) |
| std = np.sqrt(variance) |
| x = (z * std) + mean |
| x = x.clip(0, 255) |
| return x |
