def get_optimal_line_heights(
api: Api, renderer: AssRenderer
) -> T.Dict[str, float]:
TEST_LINE_COUNT = 20
VIDEO_RES_X = 100
VIDEO_RES_Y = TEST_LINE_COUNT * 300
fake_meta = AssMeta()
fake_meta.set("WrapStyle", 2)
renderer.set_source(
style_list=api.subs.styles,
event_list=api.subs.events,
meta=fake_meta,
video_resolution=(VIDEO_RES_X, VIDEO_RES_Y),
)
ret = {}
for style in api.subs.styles:
event = AssEvent(
start=0,
end=1000,
text="\\N".join(["gjMW"] * TEST_LINE_COUNT),
style=style.name,
)
_frame_width, frame_height = measure_frame_size(api, renderer, event)
line_height = frame_height / TEST_LINE_COUNT
ret[event.style] = line_height
api.log.debug(f"average height for {event.style}: {line_height}")
return ret
def measure_frame_size(
api: Api, renderer: AssRenderer, event: AssEvent
) -> T.Tuple[int, int]:
fake_event_list = AssEventList()
fake_event_list.append(copy(event))
renderer.set_source(
style_list=renderer.style_list,
event_list=fake_event_list,
meta=renderer.meta,
video_resolution=renderer.video_resolution,
)
layers = [
layer
for layer in renderer.render_raw(time=event.start)
if layer.type == 0
]
if not layers:
return (0, 0)
min_x = min(layer.dst_x for layer in layers)
min_y = min(layer.dst_y for layer in layers)
max_x = max(layer.dst_x + layer.w for layer in layers)
max_y = max(layer.dst_y + layer.h for layer in layers)
return (int((max_x - min_x) * api.video.aspect_ratio), max_y - min_y)
def get_optimal_line_heights(
api: Api, renderer: AssRenderer
) -> T.Dict[str, float]:
TEST_LINE_COUNT = 20
VIDEO_RES_X = 100
VIDEO_RES_Y = TEST_LINE_COUNT * 300
fake_meta = AssMeta()
fake_meta.set("WrapStyle", 2)
renderer.set_source(
style_list=api.subs.styles,
event_list=api.subs.events,
meta=fake_meta,
video_resolution=(VIDEO_RES_X, VIDEO_RES_Y),
)
ret = {}
for style in api.subs.styles:
event = AssEvent(
start=0,
end=1000,
text="\\N".join(["gjMW"] * TEST_LINE_COUNT),
style=style.name,
)
_frame_width, frame_height = measure_frame_size(api, renderer, event)
line_height = frame_height / TEST_LINE_COUNT
ret[event.style] = line_height
api.log.debug(f"average height for {event.style}: {line_height}")
return ret
def measure_frame_size(
api: Api, renderer: AssRenderer, event: AssEvent
) -> T.Tuple[int, int]:
if not any(style.name == event.style for style in renderer.style_list):
return (0, 0)
fake_event_list = AssEventList()
fake_event_list.append(copy(event))
renderer.set_source(
style_list=renderer.style_list,
event_list=fake_event_list,
meta=renderer.meta,
video_resolution=renderer.video_resolution,
)
layers = [
layer
for layer in renderer.render_raw(time=event.start)
if layer.type == 0
]
if not layers:
return (0, 0)
min_x = min(layer.dst_x for layer in layers)
min_y = min(layer.dst_y for layer in layers)
max_x = max(layer.dst_x + layer.w for layer in layers)
max_y = max(layer.dst_y + layer.h for layer in layers)
aspect_ratio = (
api.video.current_stream.aspect_ratio
if api.video.current_stream
else 1
)
return (int((max_x - min_x) * aspect_ratio), max_y - min_y)
def list_violations(spell_check_lang: T.Optional[str],
api: Api) -> T.Iterable[BaseResult]:
renderer = AssRenderer()
optimal_line_heights = get_optimal_line_heights(api, renderer)
renderer.set_source(
style_list=api.subs.styles,
event_list=api.subs.events,
meta=api.subs.meta,
video_resolution=(get_width(api), get_height(api)),
)
for event in api.subs.events:
yield from check_style_validity(event, api.subs.styles)
yield from check_durations(event)
yield from check_punctuation(spell_check_lang, event)
yield from check_quotes(event)
yield from check_line_continuation(event)
yield from check_ass_tags(event)
yield from check_double_words(event)
yield from check_unnecessary_breaks(event, api, renderer)
yield from check_long_line(event, api, renderer, optimal_line_heights)
def list_violations(api: Api) -> T.Iterable[BaseResult]:
renderer = AssRenderer()
optimal_line_heights = get_optimal_line_heights(api, renderer)
renderer.set_source(
style_list=api.subs.styles,
event_list=api.subs.events,
meta=api.subs.meta,
video_resolution=(get_width(api), get_height(api)),
)
for event in api.subs.events:
yield from check_style_validity(event, api.subs.styles)
yield from check_durations(event)
yield from check_punctuation(event)
yield from check_quotes(event)
yield from check_line_continuation(event)
yield from check_ass_tags(event)
yield from check_double_words(event)
yield from check_unnecessary_breaks(
event, api, renderer, optimal_line_heights
)
yield from check_long_line(event, api, renderer, optimal_line_heights)
class _StylePreview(QtWidgets.QGroupBox):
preview_text_changed = QtCore.pyqtSignal([])
def __init__(
self,
api: Api,
model: AssStylesModel,
selection_model: QtCore.QItemSelectionModel,
parent: QtWidgets.QWidget,
) -> None:
super().__init__("Preview", parent)
self._api = api
self._selection_model = selection_model
self._renderer = AssRenderer()
self._editor = QtWidgets.QPlainTextEdit()
self._editor.setPlainText(api.cfg.opt["styles"]["preview_test_text"])
self._editor.setFixedWidth(400)
self._editor.setTabChangesFocus(True)
self._editor.setFixedHeight(get_text_edit_row_height(self._editor, 2))
self._background_combobox = QtWidgets.QComboBox()
for i, path in enumerate(get_assets("style_preview_bk")):
self._background_combobox.addItem(path.name, path.resolve())
if path.name == api.cfg.opt["styles"]["preview_background"]:
self._background_combobox.setCurrentIndex(i)
self._preview_box = QtWidgets.QLabel(self)
self._preview_box.setLineWidth(1)
self._preview_box.setFrameShape(QtWidgets.QFrame.StyledPanel)
self._preview_box.setFrameShadow(QtWidgets.QFrame.Sunken)
self._preview_box.setSizePolicy(
QtWidgets.QSizePolicy.Ignored, QtWidgets.QSizePolicy.Ignored
)
layout = QtWidgets.QVBoxLayout(self)
layout.addWidget(self._editor)
layout.addWidget(self._background_combobox)
layout.addWidget(self._preview_box)
self.update_preview()
self._editor.textChanged.connect(self._on_text_change)
self._background_combobox.currentIndexChanged.connect(
self._on_background_change
)
model.dataChanged.connect(self.update_preview)
model.rowsInserted.connect(self.update_preview)
model.rowsRemoved.connect(self.update_preview)
selection_model.selectionChanged.connect(self.update_preview)
def _on_background_change(self) -> None:
self.update_preview()
self._api.cfg.opt["styles"][
"preview_background"
] = self._background_combobox.currentData().name
def _on_text_change(self) -> None:
self.preview_text_changed.emit()
self.update_preview()
self._api.cfg.opt["styles"]["preview_test_text"] = self.preview_text
@property
def preview_text(self) -> str:
return self._editor.toPlainText()
@property
def _selected_style(self) -> T.Optional[AssStyle]:
try:
idx = self._selection_model.selectedIndexes()[0].row()
except IndexError:
return None
else:
return self._api.subs.styles[idx]
def update_preview(self) -> None:
selected_style = self._selected_style
if not selected_style:
self._preview_box.clear()
return
resolution = (self._preview_box.width(), self._preview_box.height())
if resolution[0] <= 0 or resolution[1] <= 0:
self._preview_box.clear()
return
fake_style = copy(selected_style)
fake_style.name = "Default"
if (
self._api.video.current_stream
and self._api.video.current_stream.is_ready
):
fake_style.scale(
resolution[1] / self._api.video.current_stream.height
)
fake_style_list = AssStyleList()
fake_style_list.append(fake_style)
fake_event = AssEvent(
start=0,
end=1000,
text=self.preview_text.replace("\n", "\\N"),
style=fake_style.name,
)
fake_event_list = AssEventList()
fake_event_list.append(fake_event)
fake_meta = AssMeta()
image = PIL.Image.new(mode="RGBA", size=resolution)
background_path = self._background_combobox.currentData()
if background_path and background_path.exists():
background = PIL.Image.open(background_path)
for y in range(0, resolution[1], background.height):
for x in range(0, resolution[0], background.width):
image.paste(background, (x, y))
self._renderer.set_source(
fake_style_list, fake_event_list, fake_meta, resolution
)
subs_image = self._renderer.render(
time=0,
aspect_ratio=(
self._api.video.current_stream.aspect_ratio
if self._api.video.current_stream
else 1
),
)
image = PIL.Image.composite(subs_image, image, subs_image)
image = PIL.ImageQt.ImageQt(image)
image = QtGui.QImage(image)
self._preview_box.setPixmap(QtGui.QPixmap.fromImage(image))
class VideoStream(QtCore.QObject):
"""The video API."""
errored = QtCore.pyqtSignal()
changed = QtCore.pyqtSignal()
loaded = QtCore.pyqtSignal()
def __init__(
self,
threading_api: ThreadingApi,
log_api: LogApi,
subs_api: SubtitlesApi,
path: Path,
) -> None:
"""Initialize self.
:param threading_api: threading API
:param log_api: logging API
:param subs_api: subtitles API
:param path: path to the video file to load
"""
super().__init__()
self._threading_api = threading_api
self._log_api = log_api
self._subs_api = subs_api
self.uid = uuid.uuid4()
self._path = path
self._timecodes: T.List[int] = []
self._keyframes: T.List[int] = []
self._frame_rate = fractions.Fraction(0, 1)
self._aspect_ratio = fractions.Fraction(1, 1)
self._width = 0
self._height = 0
self._ass_renderer = AssRenderer()
self._source: T.Union[None, ffms2.VideoSource] = None
self._last_output_fmt: T.Any = None
self._log_api.info(f"video: loading {path}")
self._threading_api.schedule_task(
lambda: _load_video_source(self._log_api, self.uid, self._path),
self._got_source,
)
@property
def path(self) -> Path:
"""Return video source path.
:return: path
"""
return self._path
@property
def is_ready(self) -> bool:
"""Return whether if the video is loaded.
:return: whether if the video is loaded
"""
return self._source is not None
def screenshot(
self,
pts: int,
path: Path,
include_subtitles: bool,
width: T.Optional[int],
height: T.Optional[int],
) -> None:
"""Save a screenshot into specified destination.
:param pts: pts to make screenshot of
:param path: path to save the screenshot to
:param include_subtitles: whether to 'burn in' the subtitles
:param width: optional width to render to
:param height: optional height to render to
"""
if width and height:
grab_width = width
grab_height = height
elif height:
grab_width = int(self.width * height / self.height)
grab_height = height
elif width:
grab_height = int(self.height * width / self.width)
grab_width = width
else:
grab_width = self.width
grab_height = self.height
if grab_width <= 0 or grab_height <= 0:
raise ValueError("cannot take a screenshot at negative resolution")
pts = self.align_pts_to_prev_frame(pts)
idx = self.timecodes.index(pts)
frame = self.get_frame(idx, grab_width, grab_height)
image = PIL.Image.frombytes("RGB", (grab_width, grab_height), frame)
if include_subtitles:
self._ass_renderer.set_source(
self._subs_api.styles,
self._subs_api.events,
self._subs_api.meta,
(grab_width, grab_height),
)
subs_image = self._ass_renderer.render(
time=pts, aspect_ratio=self._aspect_ratio)
image = PIL.Image.composite(subs_image, image, subs_image)
image.save(str(path))
def align_pts_to_near_frame(self, pts: int) -> int:
"""Align PTS to a frame closest to given PTS.
:param pts: PTS to align
:return: aligned PTS
"""
if self.timecodes:
max_idx = len(self.timecodes) - 1
idx1 = max(
0, min(max_idx,
bisect.bisect_right(self.timecodes, pts) - 1))
idx2 = max(0, min(max_idx, bisect.bisect_left(self.timecodes,
pts)))
return min(
[self.timecodes[idx1], self.timecodes[idx2]],
key=lambda val: abs(val - pts),
)
return pts
def align_pts_to_prev_frame(self, pts: int) -> int:
"""Align PTS to a frame immediately before given PTS.
:param pts: PTS to align
:return: aligned PTS
"""
if self.timecodes:
idx = bisect.bisect_right(self.timecodes, pts) - 1
if idx >= len(self.timecodes):
return self.timecodes[-1]
if pts < self.timecodes[idx]:
return pts
return self.timecodes[idx]
return pts
def align_pts_to_next_frame(self, pts: int) -> int:
"""Align PTS to a frame immediately after given PTS.
:param pts: PTS to align
:return: aligned PTS
"""
if self.timecodes:
idx = bisect.bisect_left(self.timecodes, pts)
if idx >= len(self.timecodes):
return pts
if pts < 0:
return self.timecodes[0]
return self.timecodes[idx]
return pts
def frame_idx_from_pts(
self, pts: T.Union[float, int,
np.array]) -> T.Union[int, np.array]:
"""Get index of a frame that contains given PTS.
:param pts: PTS to search for
:return: frame index, -1 if not found
"""
ret = np.searchsorted(self.timecodes, pts, "right").astype(np.int)
ret = np.clip(ret - 1, a_min=0 if self.timecodes else -1, a_max=None)
return ret
@property
def frame_rate(self) -> fractions.Fraction:
"""Return the frame rate.
:return: video frame rate
"""
return self._frame_rate
@property
def width(self) -> int:
"""Return horizontal video resolution.
:return: video width in pixels
"""
return self._width
@property
def height(self) -> int:
"""Return vertical video resolution.
:return: video height in pixels
"""
return self._height
@property
def aspect_ratio(self) -> fractions.Fraction:
"""Return the frame aspect ratio.
:return: video frame aspect ratio
"""
return self._aspect_ratio
@property
def timecodes(self) -> T.List[int]:
"""Return video frames' PTS.
:return: video frames' PTS
"""
if not self._wait_for_source():
return []
return self._timecodes
@property
def keyframes(self) -> T.List[int]:
"""Return video keyframes' indexes.
:return: video keyframes' indexes
"""
if not self._wait_for_source():
return []
return self._keyframes
@property
def min_pts(self) -> int:
"""Return minimum video time in milliseconds.
:return: minimum PTS
"""
if not self.timecodes:
return 0
return self.timecodes[0]
@property
def max_pts(self) -> int:
"""Return maximum video time in milliseconds.
:return: maximum PTS
"""
if not self.timecodes:
return 0
return self.timecodes[-1]
def get_frame(self, frame_idx: int, width: int,
height: int) -> T.Optional[np.array]:
"""Get raw video data from the currently loaded video source.
:param frame_idx: frame number
:param width: output image width
:param height: output image height
:return: numpy image
"""
with _SAMPLER_LOCK:
if (not self._wait_for_source() or frame_idx < 0
or frame_idx >= len(self.timecodes)):
return None
assert self._source
new_output_fmt = (_PIX_FMT, width, height, ffms2.FFMS_RESIZER_AREA)
if self._last_output_fmt != new_output_fmt:
self._source.set_output_format(*new_output_fmt)
self._last_output_fmt = new_output_fmt
frame = self._source.get_frame(frame_idx)
return (frame.planes[0].reshape(
(height,
frame.Linesize[0]))[:, 0:width * 3].reshape(height, width, 3))
def _got_source(self, source: ffms2.VideoSource) -> None:
with _SAMPLER_LOCK:
self._source = source
if source is None:
self.errored.emit()
return
self._timecodes = [
int(round(pts)) for pts in source.track.timecodes
]
self._keyframes = source.track.keyframes[:]
self._timecodes.sort()
self._keyframes.sort()
self._frame_rate = fractions.Fraction(
self._source.properties.FPSNumerator,
self._source.properties.FPSDenominator,
)
self._aspect_ratio = (fractions.Fraction(
self._source.properties.SARNum,
self._source.properties.SARDen,
) if (self._source.properties.SARNum
and self._source.properties.SARDen) else fractions.Fraction(
1, 1))
frame = source.get_frame(0)
self._width = frame.EncodedWidth
self._height = int(frame.EncodedHeight / self._aspect_ratio)
self.loaded.emit()
def _wait_for_source(self) -> bool:
if self._source is None:
return False
while self._source is _LOADING:
time.sleep(0.01)
if self._source is None:
return False
return True