File size: 4,772 Bytes
3eb682b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 |
"""
Modified from https://github.com/m-bain/frozen-in-time/blob/22a91d78405ec6032fdf521ae1ff5573358e632f/base/base_dataset.py
"""
import random
import av
import decord
import torch
import numpy as np
import math
# decord.bridge.set_bridge("torch")
from decord.bridge import to_torch
def pts_to_secs(pts: int, time_base: float, start_pts: int) -> float:
"""
Converts a present time with the given time base and start_pts offset to seconds.
Returns:
time_in_seconds (float): The corresponding time in seconds.
https://github.com/facebookresearch/pytorchvideo/blob/main/pytorchvideo/data/utils.py#L54-L64
"""
if pts == math.inf:
return math.inf
return int(pts - start_pts) * time_base
def get_pyav_video_duration(video_reader):
video_stream = video_reader.streams.video[0]
video_duration = pts_to_secs(
video_stream.duration,
video_stream.time_base,
video_stream.start_time
)
return float(video_duration)
def get_frame_indices_by_fps():
pass
def get_frame_indices(num_frames, vlen, sample='rand', fix_start=None, input_fps=1, max_num_frames=-1):
if sample in ["rand", "middle"]:
acc_samples = min(num_frames, vlen)
# split the video into `acc_samples` intervals, and sample from each interval.
intervals = np.linspace(start=0, stop=vlen, num=acc_samples + 1).astype(int)
ranges = []
for idx, interv in enumerate(intervals[:-1]):
ranges.append((interv, intervals[idx + 1] - 1))
if sample == 'rand':
try:
frame_indices = [random.choice(range(x[0], x[1])) for x in ranges]
except:
frame_indices = np.random.permutation(vlen)[:acc_samples]
frame_indices.sort()
frame_indices = list(frame_indices)
elif fix_start is not None:
frame_indices = [x[0] + fix_start for x in ranges]
elif sample == 'middle':
frame_indices = [(x[0] + x[1]) // 2 for x in ranges]
else:
raise NotImplementedError
if len(frame_indices) < num_frames: # padded with last frame
padded_frame_indices = [frame_indices[-1]] * num_frames
padded_frame_indices[:len(frame_indices)] = frame_indices
frame_indices = padded_frame_indices
elif "fps" in sample: # fps0.5, sequentially sample frames at 0.5 fps
output_fps = float(sample[3:])
duration = float(vlen) / input_fps
delta = 1 / output_fps # gap between frames, this is also the clip length each frame represents
frame_seconds = np.arange(0 + delta / 2, duration + delta / 2, delta)
frame_indices = np.around(frame_seconds * input_fps).astype(int)
frame_indices = [e for e in frame_indices if e < vlen]
if max_num_frames > 0 and len(frame_indices) > max_num_frames:
frame_indices = frame_indices[:max_num_frames]
# frame_indices = np.linspace(0 + delta / 2, duration + delta / 2, endpoint=False, num=max_num_frames)
else:
raise ValueError
return frame_indices
def read_frames_av(video_path, num_frames, sample='rand', fix_start=None, max_num_frames=-1):
reader = av.open(video_path)
frames = [torch.from_numpy(f.to_rgb().to_ndarray()) for f in reader.decode(video=0)]
vlen = len(frames)
duration = get_pyav_video_duration(reader)
fps = vlen / float(duration)
frame_indices = get_frame_indices(
num_frames, vlen, sample=sample, fix_start=fix_start,
input_fps=fps, max_num_frames=max_num_frames
)
frames = torch.stack([frames[idx] for idx in frame_indices]) # (T, H, W, C), torch.uint8
frames = frames.permute(0, 3, 1, 2) # (T, C, H, W), torch.uint8
return frames, frame_indices, duration
# decord.bridge.set_bridge("torch")
def read_frames_decord(video_path, num_frames, sample='rand', fix_start=None, max_num_frames=-1):
video_reader = decord.VideoReader(video_path, num_threads=1)
vlen = len(video_reader)
fps = video_reader.get_avg_fps()
duration = vlen / float(fps)
frame_indices = get_frame_indices(
num_frames, vlen, sample=sample, fix_start=fix_start,
input_fps=fps, max_num_frames=max_num_frames
)
frames = video_reader.get_batch(frame_indices) # (T, H, W, C), torch.uint8
frames = to_torch(frames)
# try:
# print(type(frames))
# frames = frames.asnumpy()
# frames = torch.from_numpy(frames)
# except:
# print("expt", type(frames))
# pass
frames = frames.permute(0, 3, 1, 2) # (T, C, H, W), torch.uint8
return frames, frame_indices, duration
VIDEO_READER_FUNCS = {
'av': read_frames_av,
'decord': read_frames_decord
}
|