def __init__(self,
filename,
ismask=False,
has_mask=False,
audio=True,
audio_buffersize=200000,
audio_fps=44100,
audio_nbytes=2,
verbose=False):
VideoClip.__init__(self, ismask)
# Make a reader
pix_fmt = "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize=audio_buffersize,
fps=audio_fps,
nbytes=audio_nbytes)
def get_video(vid_path, color=True, size=True):
"""Get video by given video path.
Parameters
----------
image_path : string
target image absolute path
color : bool
if color is True then return color frames with BGR encoding.
if color is False then return grey scale frames.
size : bool
if size is True then return the size of the frame.
if size is False then just return the frame.
Returns
-------
frames : list
a list of frames that contains the video
size : tuple
size of the frame (optional).
"""
vid_container = FFMPEG_VideoReader(vid_path)
frames = []
for i in range(vid_container.nframes):
frame_t = vid_container.read_frame()
frame_t = cv2.cvtColor(frame_t, cv2.COLOR_RGB2BGR)
if color is False:
frame_t = cv2.cvtColor(frame_t, cv2.COLOR_BGR2GRAY)
frames.append(frame_t)
if size is True:
return frames, frames[0].shape
else:
return frames
def get_video(vid_path, color=True, size=True):
"""Get video by given video path.
Parameters
----------
image_path : string
target image absolute path
color : bool
if color is True then return color frames with BGR encoding.
if color is False then return grey scale frames.
size : bool
if size is True then return the size of the frame.
if size is False then just return the frame.
Returns
-------
frames : list
a list of frames that contains the video
size : tuple
size of the frame (optional).
"""
vid_container = FFMPEG_VideoReader(vid_path)
frames = []
for i in range(vid_container.nframes):
frame_t = vid_container.read_frame()
frame_t = cv2.cvtColor(frame_t, cv2.COLOR_RGB2BGR)
if color is False:
frame_t = cv2.cvtColor(frame_t, cv2.COLOR_BGR2GRAY)
frames.append(frame_t)
if size is True:
return frames, frames[0].shape
else:
return frames
def startpointCompare(queryVidName, dbVidName, dbStart):
"""
Startpoint compare take a frame number worth pursuing, and calculates the
average rmse value for the duration of the video starting at that point
"""
#Create the FFMPEG class variables
dbVid = FFMPEG_VideoReader(dbVidName)
queryVid = FFMPEG_VideoReader(queryVidName)
length = queryVid.nframes
# Skip to the startpoint frame
dbVid.skip_frames(dbStart)
frameD = dbVid.read_frame()
frameQ = queryVid.read_frame()
runAvg = 0
# Calculate the RMSE for each frame
for i in xrange(0, length):
runAvg += frame_rmse(frameD, frameQ)
frameQ = queryVid.read_frame()
frameD = dbVid.read_frame()
# Return the average RMSE score
return runAvg / length
def test_moviepy(infile, out_dir):
import numpy as np
import pipi
from moviepy.video.io.ffmpeg_reader import FFMPEG_VideoReader
from moviepy.video.io.ffmpeg_writer import FFMPEG_VideoWriter
chunksize = 128
vr = FFMPEG_VideoReader(infile)
w,h = vr.size
chunk = np.zeros((h, w, vr.depth, chunksize), dtype=np.uint8)
with pipi.Timer("mp read..."):
for i in range(chunksize):
frame = vr.read_frame()
chunk[...,i] = frame
vw = FFMPEG_VideoWriter(os.path.join(out_dir, "mp_ov.mkv"), (w,h), 30, codec="libx264", preset="fast", ffmpeg_params=["-crf", "0"])
with pipi.Timer("mp write..."):
for i in range(chunksize):
vw.write_frame(chunk[...,i])
vr.close()
vw.close()
def __load_set(self, set_file):
with open(set_file) as f:
lines = f.readlines()
video_list = []
text_list = []
gt_list = []
for line in lines:
line = line.strip('\n')
segs = line.split(' ')
print('=>Load Video', segs)
assert (len(segs) == 3)
segs = [os.path.join(self.root, seg) for seg in segs]
video_list.append(segs[0])
cap = FFMPEG_VideoReader(segs[0])
cap.initialize()
#video_list.append(cap)
print('Video: frames({})'.format(int(cap.nframes)))
# Load text json file
text = json.load(open(segs[1]))
# Load GT json file
gt = np.load(open(segs[2]))
print('Gt : frames({})'.format(len(gt)))
text_list.append(text)
gt_list.append(gt)
self.video_list = video_list
self.text_list = text_list
self.gt_list = gt_list
def test_moviepy(infile, out_dir):
import numpy as np
import pipi
from moviepy.video.io.ffmpeg_reader import FFMPEG_VideoReader
from moviepy.video.io.ffmpeg_writer import FFMPEG_VideoWriter
chunksize = 128
vr = FFMPEG_VideoReader(infile)
w,h = vr.size
chunk = np.zeros((h, w, vr.depth, chunksize), dtype=np.uint8)
with pipi.Timer("mp read..."):
for i in range(chunksize):
frame = vr.read_frame()
chunk[...,i] = frame
vw = FFMPEG_VideoWriter(os.path.join(out_dir, "mp_ov.mkv"), (w,h), 30, codec="libx264", preset="fast", ffmpeg_params=["-crf", "0"])
with pipi.Timer("mp write..."):
for i in range(chunksize):
vw.write_frame(chunk[...,i])
vr.close()
vw.close()
def __play_video_ffmpeg(video_path, caption, window_name='window', speed=1):
is_playing = True
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = float(cap.fps)
n_frames = cap.nframes
index = 0
while True:
if is_playing:
time_sec = index / fps
# increment by speed
index += speed
frame = cap.get_frame(time_sec)
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
frame_size = frame.shape
# resize the frame
f_width = 800
resize_factor = float(f_width) / frame_size[1]
f_height = int(frame_size[0] * resize_factor)
frame_size = (f_width, f_height)
frame = cv2.resize(src=frame,
dsize=frame_size,
interpolation=cv2.INTER_AREA)
# write caption on frame
top = int((f_height * 0.9))
text_width = cv2.getTextSize(caption, font, 1.2, 1)[0][0] + 20
cv2.rectangle(frame, (0, top - 22), (text_width, top + 10),
black_color, cv2.cv.CV_FILLED)
cv2.putText(img=frame,
text=caption,
org=(10, top),
fontFace=font,
fontScale=1.2,
color=white_color,
thickness=1,
lineType=8)
# show the frame
cv2.imshow(window_name, frame)
e = cv2.waitKey(2)
if e == 27:
break
if e == 32:
is_playing = False
print('Pause video')
# If the number of captured frames is equal to the total number of frames,we stop
if index >= n_frames:
break
else:
# toggle pause with 'space'
e = cv2.waitKey(2)
if e == 32:
is_playing = True
print('Play video')
def __init__(
self,
filename,
has_mask=False,
audio=True,
audio_buffersize=200000,
target_resolution=None,
resize_algorithm="bicubic",
audio_fps=44100,
audio_nbytes=2,
fps_source="tbr",
):
VideoClip.__init__(self)
# Make a reader
pix_fmt = "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(
filename,
pix_fmt=pix_fmt,
target_resolution=target_resolution,
resize_algo=resize_algorithm,
fps_source=fps_source,
)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.rotation = self.reader.rotation
self.filename = filename
if has_mask:
self.make_frame = lambda t: self.reader.get_frame(t)[:, :, :3]
def mask_mf(t):
return self.reader.get_frame(t)[:, :, 3] / 255.0
self.mask = VideoClip(
ismask=True, make_frame=mask_mf).set_duration(self.duration)
self.mask.fps = self.fps
else:
self.make_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos["audio_found"]:
self.audio = AudioFileClip(
filename,
buffersize=audio_buffersize,
fps=audio_fps,
nbytes=audio_nbytes,
)
def __init__(self, filename, max_frames=None, starttime=0, duration=None, asfloat=True, scale=None):
FFMPEG_VideoReader.__init__(self, filename, check_duration=True,
starttime=starttime, duration=duration)
self.width, self.height = self.size
self.gen = self._gen()
self.max_frames = max_frames
self.asfloat = asfloat
self.scale = scale
def __init__(self, filename, max_frames=None, starttime=0, duration=None, asfloat=True, scale=None):
FFMPEG_VideoReader.__init__(self, filename, check_duration=True,
starttime=starttime, duration=duration)
self.width, self.height = self.size
self.gen = self._gen()
self.max_frames = max_frames
self.asfloat = asfloat
self.scale = scale
def get_frame(video, t_frame=0.0):
"""Crop a single frame from the video to check cropping result. Stored alongside video.
:param video: Path to video
:param t_frame: position to extract example frame from, in seconds. Default: 0
:return: image
"""
clip = FFMPEG_VideoReader(str(video))
image = clip.get_frame(t_frame)
return image
def __init__(self,
filename,
has_mask=False,
audio=True,
audio_buffersize=200000,
target_resolution=None,
resize_algorithm='bicubic',
audio_fps=44100,
audio_nbytes=2,
verbose=False,
fps_source='tbr'):
VideoClip.__init__(self)
# Make a reader
pix_fmt = "rgba" if has_mask else "rgb24"
self.reader = None # need this just in case FFMPEG has issues (__del__ complains)
self.reader = FFMPEG_VideoReader(filename,
pix_fmt=pix_fmt,
target_resolution=target_resolution,
resize_algo=resize_algorithm,
fps_source=fps_source)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.rotation = self.reader.rotation
self.filename = self.reader.filename
if has_mask:
self.make_frame = lambda t: self.reader.get_frame(t)[:, :, :3]
mask_mf = lambda t: self.reader.get_frame(t)[:, :, 3] / 255.0
self.mask = (VideoClip(
ismask=True, make_frame=mask_mf).set_duration(self.duration))
self.mask.fps = self.fps
else:
self.make_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize=audio_buffersize,
fps=audio_fps,
nbytes=audio_nbytes)
def test_large_skip_frame_pos():
reader = FFMPEG_VideoReader("media/big_buck_bunny_0_30.webm")
assert reader.fps == 24
# 10 sec * 24 fps = 240 frames
frame = reader.get_frame(240 // 24)
assert reader.pos == 241
frame2 = reader.get_frame(719 / 24)
assert reader.pos == 720
# Go backwards
backwards_frame = reader.get_frame(120 // 24)
assert reader.pos == 121
def get_frames(video_path: str) -> List[np.ndarray]:
"""
Load frames from video.
:param video_path: path to video.
:return: loaded frames.
"""
video_reader = FFMPEG_VideoReader(video_path)
frames = []
for _ in tqdm(range(video_reader.nframes), desc='Getting video frames'):
frames.append(video_reader.read_frame())
return frames
def save_frames_from_vid(video_name, category_id, train_or_test, row=0):
make_dir_structure(row)
# Initialize FFMPEG_VideoReader
fvr = FFMPEG_VideoReader(filename=video_name)
fvr.initialize()
vid = os.path.split(video_name)[1]
for i in range(0, fvr.nframes):
frame_name = vid + '_' + str(i)
frame = fvr.read_frame()
imsave(os.path.join('trafficdb', 'eval_'+ str(row), train_or_test, str(category_id), frame_name + '.jpg'),
frame)
return True
def default_loader(path, frame_selector, transform=None):
video = Video(path)
frames = frame_selector.select(video.nframes, video.fps)
if transform:
state = random.getstate()
data = [_same_transform((video.get_frame(f / video.fps)), transform, state) for f in frames]
random.seed(None)
else:
data = [video.get_frame(f / video.fps) for f in frames]
if isinstance(data[0], np.ndarray):
return np.stack(data)
elif isinstance(data[0], torch.Tensor):
return torch.stack(data)
else:
raise RuntimeError()
def __init__(self, filename, ismask=False, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self, ismask)
# We store the construction parameters in case we need to make
# a copy (a 'co-reader').
self.parameters = {'filename':filename, 'ismask':ismask,
'has_mask':has_mask, 'audio':audio,
'audio_buffersize':audio_buffersize}
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio:
try:
self.audio = AudioFileClip(filename, buffersize= audio_buffersize,
fps = audio_fps, nbytes = audio_nbytes)
except:
if verbose:
print "No audio found in %s"%filename
pass
def get_frames_from_vid(video_name, category_id):
# Initialize FFMPEG_VideoReader
fvr = FFMPEG_VideoReader(filename=video_name)
fvr.initialize()
shape_for_stack = (1, fvr.size[0], fvr.size[1], fvr.depth)
img_stack = np.zeros(shape_for_stack)
for i in range(0, fvr.nframes):
frame = fvr.read_frame()
frame = frame.reshape(shape_for_stack)
img_stack = np.vstack((img_stack, frame))
img_stack = img_stack[1:]
cat_stack = np.ones((len(img_stack), 1)) * category_id
return img_stack, cat_stack
def __init__(self, filename, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self)
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
if has_mask:
self.get_frame = lambda t: self.reader.get_frame(t)[:,:,:3]
mask_gf = lambda t: self.reader.get_frame(t)[:,:,3]/255.0
self.mask = (VideoClip(ismask = True, get_frame = mask_gf)
.set_duration(self.duration))
self.mask.fps = self.fps
else:
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
def video_uniform_sample_n_frames_old(video_path, n_samples, max_dim):
"""
Sample only n frames from the video.
"""
raise Exception('Needs to add argument about resizing type')
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = cap.fps
n_frames = cap.nframes
duration = cap.duration
step = duration / (n_samples)
frames = []
for i in range(n_samples):
time_sec = i * step
frame = cap.get_frame(time_sec)
# resize frame to fit in the array, it's going to be used by caffe anyway
frame = image_utils.resize_keep_aspect_ratio_max_dim(frame, max_dim)
# frame encoded as uint and values are from 0-255
# but caffe needs float32 and values from 0-1
frame = frame.astype('float32') / float(255)
frames.append(frame)
# very important, or we'd have memory leakage
cap.__del__()
return frames
def __init__(self,
filename,
ismask=False,
has_mask=False,
audio=True,
audio_buffersize=200000,
audio_fps=44100,
audio_nbytes=2,
verbose=False):
VideoClip.__init__(self, ismask)
# We store the construction parameters in case we need to make
# a copy (a 'co-reader').
self.parameters = {
'filename': filename,
'ismask': ismask,
'has_mask': has_mask,
'audio': audio,
'audio_buffersize': audio_buffersize
}
# Make a reader
pix_fmt = "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio:
try:
self.audio = AudioFileClip(filename,
buffersize=audio_buffersize,
fps=audio_fps,
nbytes=audio_nbytes)
except:
if verbose:
print "No audio found in %s" % filename
pass
def get_regions(video_path, annot, resize_type, verbose=False):
"""
Get the frames whose numbers are given in the "annot" dictionary.. Then, for each frame get the regions as specificed in the "annot" dictionary.
Finally, return these regions.
"""
assert resize_type in ['resize', 'resize_crop', 'resize_crop_scaled']
resize_function = None
if resize_type == 'resize':
resize_function = image_utils.resize_frame
elif resize_type == 'resize_crop':
resize_function = image_utils.resize_crop
elif resize_type == 'resize_crop_scaled':
resize_function = image_utils.resize_crop_scaled
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = float(cap.fps)
n_frames = cap.nframes
duration = cap.duration
n_regions = sum([len(v) for k, v in annot.iteritems()])
frame_size = 224
bbox_resize_factor = 2
regions = np.zeros(shape=(n_regions, frame_size, frame_size, 3),
dtype='float32')
region_idx = -1
frame_nums = annot.keys()
for frame_num in frame_nums:
if (region_idx + 1) % 100 == 0 and verbose:
print(' ... reading region %d/%d' % (region_idx + 1, n_regions))
# get the frame
i = frame_num - 1
time_sec = i / fps
frame = cap.get_frame(time_sec)
# get the regions (resized) from the frame
regions_info = annot[frame_num]
for region_info in regions_info:
region_idx += 1
bbox = region_info[1:5]
bbox = np.multiply(bbox, bbox_resize_factor).astype(np.int)
x1, y1, x2, y2 = bbox
region = frame[y1:y2, x1:x2]
# resize frame to fit in the array, it's going to be used by caffe anyway
region = resize_function(region)
# frame encoded as uint and values are from 0-255, but caffe needs float32 and values from 0-1
region = region.astype('float32') / float(255)
regions[region_idx] = region
# very important, or we'd have memory leakage
cap.__del__()
return regions
def __setstate__(self, newstate):
"""
Custom unpickling
"""
# Recreate the video segment's audio and reader
newstate['reader'] = FFMPEG_VideoReader(newstate['filename'])
newstate['audio'] = AudioFileClip(newstate['filename']).subclip(
newstate['source_start_time'],
newstate['source_start_time'] + newstate['duration'])
self.__dict__.update(newstate)
def test_autorotate():
# This test requires ffmpeg >=2.7
video_file = 'media/ficus_vertical.mp4'
reader = FFMPEG_VideoReader(video_file)
assert reader.infos['video_size'] == [1920, 1080]
assert reader.infos['video_rotation'] == 90
assert reader.size == [1080, 1920]
reader.close()
reader = FFMPEG_VideoReader(video_file, ffmpeg_params=['-noautorotate'])
assert reader.size == [1920, 1080]
assert reader.rotation == 90
reader.close()
def test_sequential_frame_pos():
"""test_video.mp4 contains 5 frames at 1 fps.
Each frame is 1x1 pixels and the sequence is Red, Green, Blue, Black, White.
The rgb values are not pure due to compression.
"""
reader = FFMPEG_VideoReader("media/test_video.mp4")
assert reader.pos == 1
# Get first frame
frame_1 = reader.get_frame(0)
assert reader.pos == 1
assert np.array_equal(frame_1, [[[254, 0, 0]]])
# Get a specific sequential frame
frame_2 = reader.get_frame(1)
assert reader.pos == 2
assert np.array_equal(frame_2, [[[0, 255, 1]]])
# Get next frame. Note `.read_frame()` instead of `.get_frame()`
frame_3 = reader.read_frame()
assert reader.pos == 3
assert np.array_equal(frame_3, [[[0, 0, 255]]])
# Skip a frame
skip_frame = reader.get_frame(4)
assert reader.pos == 5
assert np.array_equal(skip_frame, [[[255, 255, 255]]])
def __init__(self,
input,
output,
width=None,
height=None,
log_level=FFMPEG_LOGLEVEL):
self._reader = FFMPEG_VideoReader(input)
self._fps = self._reader.fps
# self.cur_frame = 1
# self._frame = None
self._querier = FfmpegQuerier()
self._info = self._querier(input)
self._duration = self._querier.duration
self.draw_dict = {}
self._resize = (int(width),
int(height)) if (width and height) else None
self._loglevel = log_level
self._output = output
self._tmp_file = self._make_tmp_file(input) if self._resize else None
self._writer = FFMPEG_VideoWriter(
self._tmp_file if self.need_resize else output,
size=self._reader.size,
fps=self._reader.fps)
def __init__(self, filename, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self)
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
self.reader = reader
# Make some of the reader's attributes accessible from the clip
self.duracion = self.reader.duracion
self.fin = self.reader.duracion
self.fps = self.reader.fps
self.tamano = self.reader.tamano
if has_mask:
self.make_frame = lambda t: reader.get_frame(t)[:,:,:3]
mask_mf = lambda t: reader.get_frame(t)[:,:,3]/255.0
self.mask = (VideoClip(ismask = True, make_frame = mask_mf)
.set_duracion(self.duracion))
self.mask.fps = self.fps
else:
self.make_frame = lambda t: reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
def comparechunk(queryVidName, dbVidName, dbStart, dbEnd, thresh):
"""
This function takes the names of the files to compare, and where in the
comparison file to begin checking from. Threshold is a user determined value
that they chose qualitatively, and the GUI turned into a quantitative number
Args: path to query video, path to comparison video, start frame number,
endframe number, callback to process results, RMSE threshold (a number from
0-255, realistically should be between 10-50)
"""
# Create the FFMPEG class variables
dbVid = FFMPEG_VideoReader(dbVidName)
queryVid = FFMPEG_VideoReader(queryVidName)
# Skip to the correct frames in the video
frameQ = queryVid.get_frame(0)
dbVid.skip_frames(dbStart)
frameD = dbVid.read_frame()
scores = []
# Compare the first frame in the query video to every frame in the chunk
belowThresh = []
for i in range(dbStart, dbEnd):
score = frame_rmse(frameQ, frameD)
# Immediately look at startframes below the threshold
if frame_rmse(frameQ, frameD) < thresh:
print "Found a frame below the threshold. Scanning sequence..."
score = startpointCompare(queryVidName, dbVidName, i)
if score < thresh and score is not None:
scores.append({
"Video Name": dbVidName,
"Timestamp": secondsToTimestamp(i / dbVid.fps),
"Frame Number": i,
"Score": score
})
return scores
else:
print "A sequence had a poor score of", score, ". Moving on..."
frameD = dbVid.read_frame()
return scores
def test_seeking_beyond_file_end():
reader = FFMPEG_VideoReader("media/test_video.mp4")
frame_1 = reader.get_frame(0)
with pytest.warns(UserWarning, match="Using the last valid frame instead"):
end_of_file_frame = reader.get_frame(5)
assert np.array_equal(frame_1, end_of_file_frame)
assert reader.pos == 6
# Try again with a jump larger than 100 frames
# (which triggers different behaivour in `.get_frame()`
reader = FFMPEG_VideoReader("media/big_buck_bunny_0_30.webm")
frame_1 = reader.get_frame(0)
with pytest.warns(UserWarning, match="Using the last valid frame instead"):
end_of_file_frame = reader.get_frame(30)
assert np.array_equal(frame_1, end_of_file_frame)
assert reader.pos == 30 * 24 + 1
def video_uniform_sample_and_save_old(spf,
video_path,
frames_path,
image_name_format,
resize_type,
verbose=False):
if resize_type is not None:
assert resize_type in ['resize', 'resize_crop', 'resize_crop_scaled']
resize_function = None
if resize_type == 'resize':
resize_function = image_utils.resize_frame
elif resize_type == 'resize_crop':
resize_function = image_utils.resize_crop
elif resize_type == 'resize_crop_scaled':
resize_function = image_utils.resize_crop_scaled
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = cap.fps
n_frames = cap.nframes
duration = cap.duration
n_samples = int(duration / float(spf))
# check if no samples because the video duration is less than spf
# then at least, get 1 frame of the video
if n_samples == 0:
n_samples = 1
for i in range(n_samples):
num = i + 1
if verbose:
print(' ... reading frame %d/%d' % (num, n_samples))
time_sec = i * spf
frame = cap.get_frame(time_sec)
if resize_type is not None:
# resize frame to fit in the array, it's going to be used by caffe anyway
frame = resize_function(frame)
frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
image_name = image_name_format % (num, )
frame_path = os.path.join(frames_path, image_name)
cv2.imwrite(frame_path, frame)
# very important, or we'd have memory leakage
cap.close()
return fps, n_frames, duration
def video_save_frames_specific_duration(action_num,
video_num,
video_path,
frames_root_pathes,
start_stop_sec,
image_name_format,
verbose=False):
assert len(frames_root_pathes) == len(start_stop_sec)
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = float(cap.fps)
duration_sec = cap.duration
img_dim = 224
start_stop_sec = np.array(start_stop_sec)
for i, s_s_sec in enumerate(start_stop_sec):
start_sec, stop_sec = s_s_sec
frame_root_path = frames_root_pathes[i]
# offset of starting/stopping the action
sec_offset = 0.25
start_idx = int((start_sec + sec_offset) * fps)
stop_idx = int((stop_sec + sec_offset) * fps) + 1
if verbose:
print('action, video: %d, %d' % (action_num, video_num))
print('%d/%d' % (start_sec, stop_sec))
print('%d/%d' % (start_idx, stop_idx))
for idx_frame in range(start_idx, stop_idx):
time_sec = idx_frame / fps
if verbose and idx_frame % 100 == 0:
print('... time_sec, frame: %d/%d' % (time_sec, idx_frame))
frame = cap.get_frame(time_sec)
frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
frame = image_utils.resize_crop(frame,
target_width=img_dim,
target_height=img_dim)
image_name = image_name_format % (idx_frame, )
frame_path = os.path.join(frame_root_path, image_name)
cv2.imwrite(frame_path, frame)
# very important, or we'd have memory leakage
cap.__del__()
def get_video_info(video_path):
# video_fps, video_n_frames, video_duration = video_utils.
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = cap.fps
n_frames = cap.nframes
duration = cap.duration
cap.close()
del cap
return fps, n_frames, duration
def __init__(self, filename, has_mask=False,
audio=True, audio_buffersize = 200000,
target_resolution=None, resize_algorithm='bicubic',
audio_fps=44100, audio_nbytes=2, verbose=False,
fps_source='tbr'):
VideoClip.__init__(self)
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = None # need this just in case FFMPEG has issues (__del__ complains)
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt,
target_resolution=target_resolution,
resize_algo=resize_algorithm,
fps_source=fps_source)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.rotation = self.reader.rotation
self.filename = self.reader.filename
if has_mask:
self.make_frame = lambda t: self.reader.get_frame(t)[:,:,:3]
mask_mf = lambda t: self.reader.get_frame(t)[:,:,3]/255.0
self.mask = (VideoClip(ismask = True, make_frame = mask_mf)
.set_duration(self.duration))
self.mask.fps = self.fps
else:
self.make_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
def video_uniform_sampling(spf,
video_path,
resize_type,
is_local,
verbose=False):
assert resize_type in ['resize', 'resize_crop', 'resize_crop_scaled']
resize_function = None
if resize_type == 'resize':
resize_function = image_utils.resize_frame
elif resize_type == 'resize_crop':
resize_function = image_utils.resize_crop
elif resize_type == 'resize_crop_scaled':
resize_function = image_utils.resize_crop_scaled
cap = FFMPEG_VideoReader(video_path, False)
cap.initialize()
fps = cap.fps
n_frames = cap.nframes
duration = cap.duration
n_samples = int(duration / float(spf))
# check if no samples because the video duration is less than spf
# then at least, get 1 frame of the video
if n_samples == 0:
n_samples = 1
frame_size = 224
frames = np.zeros(shape=(n_samples, frame_size, frame_size, 3),
dtype='float32')
for i in range(n_samples):
num = i + 1
if num % 100 == 0 and verbose:
print(' ... reading frame %d/%d' % (num, n_samples))
time_sec = i * spf
frame = cap.get_frame(time_sec)
# resize frame to fit in the array, it's going to be used by caffe anyway
frame = resize_function(frame)
# frame encoded as uint and values are from 0-255
# but caffe needs float32 and values from 0-1
frame = frame.astype('float32') / float(255)
frames[i] = frame
# very important, or we'd have memory leakage
cap.__del__()
return frames, fps, n_frames, duration
def __init__(self, swf_mp4, grf_mp4, chat_mp4, output, fps=10):
self._swf_mp4 = swf_mp4
self._grf_mp4 = grf_mp4
self._chat_mp4 = chat_mp4
self._fq = FfmpegQuerier()
self._swf_reader = FFMPEG_VideoReader(swf_mp4)
self._grf_reader = FFMPEG_VideoReader(grf_mp4)
self._chat_reader = FFMPEG_VideoReader(chat_mp4)
self._output = output
self._fps = fps
self._output_size = self.cal_output_size()
self._writer = FFMPEG_VideoWriter(output,
self._output_size,
fps,
audiofile=grf_mp4)
def __init__(self, filename, ismask=False, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self, ismask)
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt,print_infos=verbose)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
class VideoFileClip(VideoClip):
"""
A video clip originating from a movie file. For instance: ::
>>> clip = VideoFileClip("myHolidays.mp4")
>>> clip.close()
>>> with VideoFileClip("myMaskVideo.avi") as clip2:
>>> pass # Implicit close called by contex manager.
Parameters
------------
filename:
The name of the video file. It can have any extension supported
by ffmpeg: .ogv, .mp4, .mpeg, .avi, .mov etc.
has_mask:
Set this to 'True' if there is a mask included in the videofile.
Video files rarely contain masks, but some video codecs enable
that. For istance if you have a MoviePy VideoClip with a mask you
can save it to a videofile with a mask. (see also
``VideoClip.write_videofile`` for more details).
audio:
Set to `False` if the clip doesn't have any audio or if you do not
wish to read the audio.
target_resolution:
Set to (desired_height, desired_width) to have ffmpeg resize the frames
before returning them. This is much faster than streaming in high-res
and then resizing. If either dimension is None, the frames are resized
by keeping the existing aspect ratio.
resize_algorithm:
The algorithm used for resizing. Default: "bicubic", other popular
options include "bilinear" and "fast_bilinear". For more information, see
https://ffmpeg.org/ffmpeg-scaler.html
fps_source:
The fps value to collect from the metadata. Set by default to 'tbr', but
can be set to 'fps', which may be helpful if importing slow-motion videos
that get messed up otherwise.
Attributes
-----------
filename:
Name of the original video file.
fps:
Frames per second in the original file.
Read docs for Clip() and VideoClip() for other, more generic, attributes.
Lifetime
--------
Note that this creates subprocesses and locks files. If you construct one of these instances, you must call
close() afterwards, or the subresources will not be cleaned up until the process ends.
If copies are made, and close() is called on one, it may cause methods on the other copies to fail.
"""
def __init__(self, filename, has_mask=False,
audio=True, audio_buffersize = 200000,
target_resolution=None, resize_algorithm='bicubic',
audio_fps=44100, audio_nbytes=2, verbose=False,
fps_source='tbr'):
VideoClip.__init__(self)
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt,
target_resolution=target_resolution,
resize_algo=resize_algorithm,
fps_source=fps_source)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.rotation = self.reader.rotation
self.filename = self.reader.filename
if has_mask:
self.make_frame = lambda t: self.reader.get_frame(t)[:,:,:3]
mask_mf = lambda t: self.reader.get_frame(t)[:,:,3]/255.0
self.mask = (VideoClip(ismask = True, make_frame = mask_mf)
.set_duration(self.duration))
self.mask.fps = self.fps
else:
self.make_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
def close(self):
""" Close the internal reader. """
if self.reader:
self.reader.close()
self.reader = None
try:
if self.audio:
self.audio.close()
self.audio = None
except AttributeError:
pass
class VideoFileClip(VideoClip):
"""
A video clip originating from a movie file. For instance: ::
>>> clip = VideofileClip("myHolidays.mp4")
>>> clip2 = VideofileClip("myMaskVideo.avi")
Parameters
------------
filename:
The name of the video file. It can have any extension supported
by ffmpeg: .ogv, .mp4, .mpeg, .avi, .mov etc.
has_mask:
Set this to 'True' if there is a mask included in the videofile.
Video files rarely contain masks, but some video codecs enable
that. For istance if you have a MoviePy VideoClip with a mask you
can save it to a videofile with a mask. (see also
``VideoClip.to_videofile`` for more details).
audio:
Set to `False` if the clip doesn't have any audio or if you do not
wish to read the audio.
Attributes
-----------
filename:
Name of the original video file.
fps:
Frames per second in the original file.
"""
def __init__(self, filename, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self)
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
if has_mask:
self.get_frame = lambda t: self.reader.get_frame(t)[:,:,:3]
mask_gf = lambda t: self.reader.get_frame(t)[:,:,3]/255.0
self.mask = (VideoClip(ismask = True, get_frame = mask_gf)
.set_duration(self.duration))
self.mask.fps = self.fps
else:
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio and self.reader.infos['audio_found']:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
class VideoFileClip(VideoClip):
"""
A video clip originating from a movie file. For instance: ::
>>> clip = VideofileClip("myHolidays.mp4")
>>> clip2 = VideofileClip("myMaskVideo.avi",ismask = True)
Parameters
------------
filename:
The name of the video file. It can have any extension supported
by ffmpeg: .ogv, .mp4, .mpeg, .avi, .mov etc.
ismask:
Set this to `True` if the clip is going to be used as a mask.
has_mask:
Set this to 'True' if there is a mask included in the videofile.
Video files rarely contain masks, but some video codecs enable
that. For istance if you have a MoviePy VideoClip with a mask you
can save it to a videofile with a mask. (see also
``VideoClip.to_videofile`` for more details).
audio:
Set to `False` if the clip doesn't have any audio or if you do not
wish to read the audio.
Attributes
-----------
filename:
Name of the original video file.
fps:
Frames per second in the original file.
"""
def __init__(self, filename, ismask=False, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self, ismask)
# We store the construction parameters in case we need to make
# a copy (a 'co-reader').
self.parameters = {'filename':filename, 'ismask':ismask,
'has_mask':has_mask, 'audio':audio,
'audio_buffersize':audio_buffersize}
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.end = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio:
self.audio = AudioFileClip(filename,
buffersize= audio_buffersize,
fps = audio_fps,
nbytes = audio_nbytes)
def coreader(self, audio=True):
"""
Returns a copy of the AudioFileClip with an autonomous reader.
Maybe REMOVED in the coming versions. Wait and see. ::
>>> clip = VideofileClip("myHolidays.mp4", ismask=True)
>>> clip2 = VideofileClip("myHolidays.mp4", ismask=True)
is equivalent to ::
>>> clip = VideofileClip("myHolidays.mp4", ismask=True)
>>> clip2 = clip.coreader()
"""
return VideoFileClip(**self.parameters)
class VideoFileClip(VideoClip):
"""
A video clip originating from a movie file. For instance:
>>> clip = VideofileClip("myHolidays.mp4")
>>> clip2 = VideofileClip("myMaskVideo.avi",ismask = True)
:param filename: Any video file: .ogv, .mp4, .mpeg, .avi, .mov etc.
:param ismask: `True` if the clip is a mask.
:param has_mask: 'True' if there is a mask included in the videofile.
Commonly, video files don't have mask, but you can save the mask
for the videos that you make with MoviePy (see the doc from
``VideoClip.to_videofile`` for more details).
:param audio: If `True`, then the audio is extracted from the video
file, in wav format, and it attributed to the clip.
:ivar filename: Name of the original video file
:ivar fps: Frames per second in the original file.
"""
def __init__(self, filename, ismask=False, has_mask=False,
audio=True, audio_buffersize = 200000,
audio_fps=44100, audio_nbytes=2, verbose=False):
VideoClip.__init__(self, ismask)
# We store the construction parameters in case we need to make
# a copy (a 'co-reader').
self.parameters = {'filename':filename, 'ismask':ismask,
'has_mask':has_mask, 'audio':audio,
'audio_buffersize':audio_buffersize}
# Make a reader
pix_fmt= "rgba" if has_mask else "rgb24"
self.reader = FFMPEG_VideoReader(filename, pix_fmt=pix_fmt)
# Make some of the reader's attributes accessible from the clip
self.duration = self.reader.duration
self.fps = self.reader.fps
self.size = self.reader.size
self.get_frame = lambda t: self.reader.get_frame(t)
# Make a reader for the audio, if any.
if audio:
try:
self.audio = AudioFileClip(filename, buffersize= audio_buffersize,
fps = audio_fps, nbytes = audio_nbytes)
except:
if verbose:
print "No audio found in %s"%filename
pass
def coreader(self, audio=True):
""" Returns a copy of the AudioFileClip, i.e. a new entrance point
to the video file. Use copy when you have different clips
watching the same video file at different times. """
return VideoFileClip(**self.parameters)
frame = gui.resize(frame, (400, 300), ratio_keep=True)
print frame.shape
if option == "test-dict-compare":
para_dict_old = {}
para_dict_old["a"] = 1
para_dict_old["b"] = 2
para_dict_new = {}
para_dict_new["a"] = 1
para_dict_new["b"] = 2
print retina.compare_para_dict(para_dict_old, para_dict_new)
if option == "test-setup-function":
eye = retina.init_retina((300, 200))
print type(eye.setupOPLandIPLParvoChannel)
print type(eye.setupIPLMagnoChannel)
print eye.getInputSize()
if option == "test-movie-py":
video = FFMPEG_VideoReader("./simretina/retina-data/HorseRiding.avi")
frame = video.read_frame()
for i in range(video.nframes):
frame = video.read_frame()
cv2.imshow("test", frame)
cv2.waitKey(0)
