Skip to content

ScreenGear API References

ScreenGear API usage examples can be found here ➶

ScreenGear API parameters are explained here ➶

ScreenGear is designed exclusively for targeting rapid Screencasting Capabilities, which means it can grab frames from your monitor in real-time, either by defining an area on the computer screen or full-screen, at the expense of inconsiderable latency. ScreenGear also seamlessly support frame capturing from multiple monitors as well as supports multiple backends.

ScreenGear API implements a multi-threaded wrapper around dxcam, pyscreenshot, python-mss python library, and also flexibly supports its internal parameter.

Source code in vidgear/gears/screengear.py
class ScreenGear:
    """
    ScreenGear is designed exclusively for targeting rapid Screencasting Capabilities, which means it can
    grab frames from your monitor in real-time, either by defining an area on the computer screen or full-screen,
    at the expense of inconsiderable latency. ScreenGear also seamlessly support frame capturing from multiple
    monitors as well as supports multiple backends.

    ScreenGear API implements a multi-threaded wrapper around dxcam, pyscreenshot, python-mss python library,
    and also flexibly supports its internal parameter.
    """

    def __init__(
        self, monitor=None, backend=None, colorspace=None, logging=False, **options
    ):
        """
        This constructor method initializes the object state and attributes of the ScreenGear class.

        Parameters:
            monitor (int): enables `mss` backend and sets the index of the monitor screen.
            backend (str): select suitable backend for extracting frames.
            colorspace (str): selects the colorspace of the input stream.
            logging (bool): enables/disables logging.
            options (dict): provides the flexibility to easily alter backend library parameters. Such as, manually set the dimensions of capture screen area etc.
        """
        # enable logging if specified
        self.__logging = logging if isinstance(logging, bool) else False

        # print current version
        logcurr_vidgear_ver(logging=self.__logging)

        # create instances for the user-defined monitor
        self.__monitor_instance = None
        self.__backend = None

        # validate monitor instance
        assert (
            monitor is None or monitor and isinstance(monitor, (int, tuple))
        ), "[ScreenGear:ERROR] :: Invalid `monitor` value detected!"

        # initialize backend
        if backend and monitor is None:
            self.__backend = backend.lower().strip()
        else:
            # enforce `dxcam` for Windows machines if undefined (or monitor is defined)
            self.__backend = (
                "dxcam" if platform.system() == "Windows" and dxcam else None
            )

        # initiate screen dimension handler
        screen_dims = {}
        # reformat proper mss dict and assign to screen dimension handler
        screen_dims = {
            k.strip(): v
            for k, v in options.items()
            if k.strip() in ["top", "left", "width", "height"]
        }
        # check whether user-defined dimensions are provided
        if screen_dims and len(screen_dims) == 4:
            key_order = (
                ("top", "left", "width", "height")
                if self.__backend != "dxcam"
                else ("left", "top", "width", "height")
            )
            screen_dims = OrderedDict((k, screen_dims[k]) for k in key_order)
            self.__logging and logger.debug(
                "Setting Capture-Area dimensions: {}".format(json.dumps(screen_dims))
            )
        else:
            screen_dims.clear()

        # handle backends
        if self.__backend == "dxcam":
            # get target fps in case of DXcam
            self.__target_fps = options.pop("dxcam_target_fps", 0)
            if self.__target_fps and isinstance(self.__target_fps, (int, float)):
                # set values
                self.__target_fps = int(self.__target_fps)
                self.__logging and logger.debug(
                    "Setting Target FPS: {}".format(self.__target_fps)
                )
            else:
                # defaults to 0fps
                self.__target_fps = 0
            # check if platform is windows
            assert (
                platform.system() == "Windows"
            ), "`dxcam` backend is only available for Windows Machines."
            # verify monitor values if tuple
            assert (
                monitor is None
                or isinstance(monitor, int)
                or (
                    isinstance(monitor, tuple)
                    and len(monitor) == 2
                    and all(isinstance(x, int) for x in monitor)
                )
            ), "For dxcam` backend, monitor` tuple value must be format `int` or `(int, int)` only."
            # raise error(s) for critical Class imports
            import_dependency_safe("dxcam" if dxcam is None else "")
            if monitor is None:
                self.__capture_object = dxcam.create(
                    region=tuple(screen_dims.values()) if screen_dims else None
                )
            else:
                self.__capture_object = (
                    dxcam.create(
                        device_idx=monitor[0],
                        output_idx=monitor[1],
                        region=tuple(screen_dims.values()) if screen_dims else None,
                    )
                    if isinstance(monitor, tuple)
                    else dxcam.create(
                        device_idx=monitor,
                        region=tuple(screen_dims.values()) if screen_dims else None,
                    )
                )
        else:
            if monitor is None:
                # raise error(s) for critical Class imports
                import_dependency_safe("pyscreenshot" if pysct is None else "")
                # reset backend if not provided
                self.__backend = "pil" if self.__backend is None else self.__backend
                # check if valid backend
                assert (
                    self.__backend in pysct.backends()
                ), "Unsupported backend {} provided!".format(backend)
                # create capture object
                self.__capture_object = pysct
            else:
                # monitor value must be integer
                assert monitor and isinstance(
                    monitor, int
                ), "[ScreenGear:ERROR] :: Invalid `monitor` value must be integer with mss backend."
                # raise error(s) for critical Class imports
                import_dependency_safe(
                    "from mss import mss" if mss is None else "", pkg_name="mss"
                )
                # create capture object
                self.__capture_object = mss()
                self.__backend and logger.warning(
                    "Backends are disabled for Monitor Indexing(monitor>=0)!"
                )
                self.__monitor_instance = self.__capture_object.monitors[monitor]

        # log backend
        self.__backend and self.__logging and logger.debug(
            "Setting Backend: {}".format(self.__backend.upper())
        )

        # assigns special parameter to global variable and clear
        # separately handle colorspace value to int conversion
        if colorspace:
            self.color_space = capPropId(colorspace.strip())
            self.__logging and not (self.color_space is None) and logger.debug(
                "Enabling `{}` colorspace for this video stream!".format(
                    colorspace.strip()
                )
            )
        else:
            self.color_space = None

        # initialize mss capture instance
        self.__mss_capture_instance = None
        try:
            if self.__backend == "dxcam":
                # extract global frame from instance
                self.frame = self.__capture_object.grab()
            else:
                if self.__monitor_instance is None:
                    if screen_dims:
                        self.__mss_capture_instance = tuple(screen_dims.values())
                    # extract global frame from instance
                    self.frame = np.asanyarray(
                        self.__capture_object.grab(
                            bbox=self.__mss_capture_instance,
                            childprocess=False,
                            backend=self.__backend,
                        )
                    )
                else:
                    if screen_dims:
                        self.__mss_capture_instance = {
                            "top": self.__monitor_instance["top"] + screen_dims["top"],
                            "left": self.__monitor_instance["left"]
                            + screen_dims["left"],
                            "width": screen_dims["width"],
                            "height": screen_dims["height"],
                            "mon": monitor,
                        }
                    else:
                        self.__mss_capture_instance = (
                            self.__monitor_instance  # otherwise create instance from monitor
                        )
                    # extract global frame from instance
                    self.frame = np.asanyarray(
                        self.__capture_object.grab(self.__mss_capture_instance)
                    )
            # convert to bgr frame if applicable
            self.frame = (
                self.frame[:, :, ::-1]
                if self.__backend == "dxcam" or not (pysct is None)
                else self.frame
            )
            # render colorspace if defined
            if not (self.frame is None) and not (self.color_space is None):
                self.frame = cv2.cvtColor(self.frame, self.color_space)
        except Exception as e:
            if isinstance(e, ScreenShotError):
                # otherwise catch and log errors
                self.__logging and logger.exception(
                    self.__capture_object.get_error_details()
                )
                raise ValueError(
                    "[ScreenGear:ERROR] :: ScreenShotError caught, Wrong dimensions passed to python-mss, Kindly Refer Docs!"
                )
            else:
                raise SystemError(
                    "[ScreenGear:ERROR] :: Unable to grab any instance on this system, Are you running headless?"
                )
        # thread initialization
        self.__thread = None
        # initialize termination flag
        self.__terminate = Event()

    def start(self):
        """
        Launches the internal *Threaded Frames Extractor* daemon

        **Returns:** A reference to the ScreenGear class object.
        """
        self.__thread = Thread(target=self.__update, name="ScreenGear", args=())
        self.__thread.daemon = True
        self.__thread.start()
        if self.__backend == "dxcam":
            self.__capture_object.start(
                target_fps=self.__target_fps,
                video_mode=True,
            )
            self.__logging and self.__target_fps and logger.debug(
                "Targeting FPS: {}".format(self.__target_fps)
            )
        return self

    def __update(self):
        """
        A **Threaded Frames Extractor**, that keep iterating frames from `mss` API to a internal monitored deque,
        until the thread is terminated, or frames runs out.
        """
        # initialize frame variable
        frame = None
        # keep looping infinitely until the thread is terminated
        while not self.__terminate.is_set():
            try:
                if self.__backend == "dxcam":
                    # extract global frame from instance
                    frame = self.__capture_object.get_latest_frame()
                else:
                    if self.__monitor_instance:
                        frame = np.asanyarray(
                            self.__capture_object.grab(self.__mss_capture_instance)
                        )
                    else:
                        frame = np.asanyarray(
                            self.__capture_object.grab(
                                bbox=self.__mss_capture_instance,
                                childprocess=False,
                                backend=self.__backend,
                            )
                        )
                # check if valid frame
                assert not (
                    frame is None or np.shape(frame) == ()
                ), "[ScreenGear:ERROR] :: Failed to retrieve valid frame!"
                # convert to bgr frame if applicable
                frame = (
                    frame[:, :, ::-1]
                    if self.__backend == "dxcam" or not (pysct is None)
                    else frame
                )
            except Exception as e:
                if isinstance(e, ScreenShotError):
                    raise RuntimeError(self.__capture_object.get_error_details())
                else:
                    logger.exception(str(e))
                self.__terminate.set()
                continue

            if not (self.color_space is None):
                # apply colorspace to frames
                color_frame = None
                try:
                    color_frame = cv2.cvtColor(frame, self.color_space)
                except Exception as e:
                    # Catch if any error occurred
                    color_frame = None
                    self.color_space = None
                    self.__logging and logger.exception(str(e))
                    logger.warning("Assigned colorspace value is invalid. Discarding!")
                self.frame = color_frame if not (color_frame is None) else frame
            else:
                self.frame = frame

        # indicate immediate termination
        self.__terminate.set()

        # finally release mss resources
        if self.__monitor_instance:
            self.__capture_object.close()
        if self.__backend == "dxcam":
            self.__capture_object.stop()
            del self.__capture_object

    def read(self):
        """
        Extracts frames synchronously from monitored deque, while maintaining a fixed-length frame buffer in the memory,
        and blocks the thread if the deque is full.

        **Returns:** A n-dimensional numpy array.
        """
        # return the frame
        return self.frame

    def stop(self):
        """
        Safely terminates the thread, and release the resources.
        """
        self.__logging and logger.debug("Terminating ScreenGear Processes.")

        # indicate that the thread should be terminate
        self.__terminate.set()

        # wait until stream resources are released (producer thread might be still grabbing frame)
        not (self.__thread is None) and self.__thread.join()

__init__(self, monitor=None, backend=None, colorspace=None, logging=False, **options) special

This constructor method initializes the object state and attributes of the ScreenGear class.

Parameters:

Name Type Description Default
monitor int

enables mss backend and sets the index of the monitor screen.

None
backend str

select suitable backend for extracting frames.

None
colorspace str

selects the colorspace of the input stream.

None
logging bool

enables/disables logging.

False
options dict

provides the flexibility to easily alter backend library parameters. Such as, manually set the dimensions of capture screen area etc.

{}
Source code in vidgear/gears/screengear.py
def __init__(
    self, monitor=None, backend=None, colorspace=None, logging=False, **options
):
    """
    This constructor method initializes the object state and attributes of the ScreenGear class.

    Parameters:
        monitor (int): enables `mss` backend and sets the index of the monitor screen.
        backend (str): select suitable backend for extracting frames.
        colorspace (str): selects the colorspace of the input stream.
        logging (bool): enables/disables logging.
        options (dict): provides the flexibility to easily alter backend library parameters. Such as, manually set the dimensions of capture screen area etc.
    """
    # enable logging if specified
    self.__logging = logging if isinstance(logging, bool) else False

    # print current version
    logcurr_vidgear_ver(logging=self.__logging)

    # create instances for the user-defined monitor
    self.__monitor_instance = None
    self.__backend = None

    # validate monitor instance
    assert (
        monitor is None or monitor and isinstance(monitor, (int, tuple))
    ), "[ScreenGear:ERROR] :: Invalid `monitor` value detected!"

    # initialize backend
    if backend and monitor is None:
        self.__backend = backend.lower().strip()
    else:
        # enforce `dxcam` for Windows machines if undefined (or monitor is defined)
        self.__backend = (
            "dxcam" if platform.system() == "Windows" and dxcam else None
        )

    # initiate screen dimension handler
    screen_dims = {}
    # reformat proper mss dict and assign to screen dimension handler
    screen_dims = {
        k.strip(): v
        for k, v in options.items()
        if k.strip() in ["top", "left", "width", "height"]
    }
    # check whether user-defined dimensions are provided
    if screen_dims and len(screen_dims) == 4:
        key_order = (
            ("top", "left", "width", "height")
            if self.__backend != "dxcam"
            else ("left", "top", "width", "height")
        )
        screen_dims = OrderedDict((k, screen_dims[k]) for k in key_order)
        self.__logging and logger.debug(
            "Setting Capture-Area dimensions: {}".format(json.dumps(screen_dims))
        )
    else:
        screen_dims.clear()

    # handle backends
    if self.__backend == "dxcam":
        # get target fps in case of DXcam
        self.__target_fps = options.pop("dxcam_target_fps", 0)
        if self.__target_fps and isinstance(self.__target_fps, (int, float)):
            # set values
            self.__target_fps = int(self.__target_fps)
            self.__logging and logger.debug(
                "Setting Target FPS: {}".format(self.__target_fps)
            )
        else:
            # defaults to 0fps
            self.__target_fps = 0
        # check if platform is windows
        assert (
            platform.system() == "Windows"
        ), "`dxcam` backend is only available for Windows Machines."
        # verify monitor values if tuple
        assert (
            monitor is None
            or isinstance(monitor, int)
            or (
                isinstance(monitor, tuple)
                and len(monitor) == 2
                and all(isinstance(x, int) for x in monitor)
            )
        ), "For dxcam` backend, monitor` tuple value must be format `int` or `(int, int)` only."
        # raise error(s) for critical Class imports
        import_dependency_safe("dxcam" if dxcam is None else "")
        if monitor is None:
            self.__capture_object = dxcam.create(
                region=tuple(screen_dims.values()) if screen_dims else None
            )
        else:
            self.__capture_object = (
                dxcam.create(
                    device_idx=monitor[0],
                    output_idx=monitor[1],
                    region=tuple(screen_dims.values()) if screen_dims else None,
                )
                if isinstance(monitor, tuple)
                else dxcam.create(
                    device_idx=monitor,
                    region=tuple(screen_dims.values()) if screen_dims else None,
                )
            )
    else:
        if monitor is None:
            # raise error(s) for critical Class imports
            import_dependency_safe("pyscreenshot" if pysct is None else "")
            # reset backend if not provided
            self.__backend = "pil" if self.__backend is None else self.__backend
            # check if valid backend
            assert (
                self.__backend in pysct.backends()
            ), "Unsupported backend {} provided!".format(backend)
            # create capture object
            self.__capture_object = pysct
        else:
            # monitor value must be integer
            assert monitor and isinstance(
                monitor, int
            ), "[ScreenGear:ERROR] :: Invalid `monitor` value must be integer with mss backend."
            # raise error(s) for critical Class imports
            import_dependency_safe(
                "from mss import mss" if mss is None else "", pkg_name="mss"
            )
            # create capture object
            self.__capture_object = mss()
            self.__backend and logger.warning(
                "Backends are disabled for Monitor Indexing(monitor>=0)!"
            )
            self.__monitor_instance = self.__capture_object.monitors[monitor]

    # log backend
    self.__backend and self.__logging and logger.debug(
        "Setting Backend: {}".format(self.__backend.upper())
    )

    # assigns special parameter to global variable and clear
    # separately handle colorspace value to int conversion
    if colorspace:
        self.color_space = capPropId(colorspace.strip())
        self.__logging and not (self.color_space is None) and logger.debug(
            "Enabling `{}` colorspace for this video stream!".format(
                colorspace.strip()
            )
        )
    else:
        self.color_space = None

    # initialize mss capture instance
    self.__mss_capture_instance = None
    try:
        if self.__backend == "dxcam":
            # extract global frame from instance
            self.frame = self.__capture_object.grab()
        else:
            if self.__monitor_instance is None:
                if screen_dims:
                    self.__mss_capture_instance = tuple(screen_dims.values())
                # extract global frame from instance
                self.frame = np.asanyarray(
                    self.__capture_object.grab(
                        bbox=self.__mss_capture_instance,
                        childprocess=False,
                        backend=self.__backend,
                    )
                )
            else:
                if screen_dims:
                    self.__mss_capture_instance = {
                        "top": self.__monitor_instance["top"] + screen_dims["top"],
                        "left": self.__monitor_instance["left"]
                        + screen_dims["left"],
                        "width": screen_dims["width"],
                        "height": screen_dims["height"],
                        "mon": monitor,
                    }
                else:
                    self.__mss_capture_instance = (
                        self.__monitor_instance  # otherwise create instance from monitor
                    )
                # extract global frame from instance
                self.frame = np.asanyarray(
                    self.__capture_object.grab(self.__mss_capture_instance)
                )
        # convert to bgr frame if applicable
        self.frame = (
            self.frame[:, :, ::-1]
            if self.__backend == "dxcam" or not (pysct is None)
            else self.frame
        )
        # render colorspace if defined
        if not (self.frame is None) and not (self.color_space is None):
            self.frame = cv2.cvtColor(self.frame, self.color_space)
    except Exception as e:
        if isinstance(e, ScreenShotError):
            # otherwise catch and log errors
            self.__logging and logger.exception(
                self.__capture_object.get_error_details()
            )
            raise ValueError(
                "[ScreenGear:ERROR] :: ScreenShotError caught, Wrong dimensions passed to python-mss, Kindly Refer Docs!"
            )
        else:
            raise SystemError(
                "[ScreenGear:ERROR] :: Unable to grab any instance on this system, Are you running headless?"
            )
    # thread initialization
    self.__thread = None
    # initialize termination flag
    self.__terminate = Event()

read(self)

Extracts frames synchronously from monitored deque, while maintaining a fixed-length frame buffer in the memory, and blocks the thread if the deque is full.

Returns: A n-dimensional numpy array.

Source code in vidgear/gears/screengear.py
def read(self):
    """
    Extracts frames synchronously from monitored deque, while maintaining a fixed-length frame buffer in the memory,
    and blocks the thread if the deque is full.

    **Returns:** A n-dimensional numpy array.
    """
    # return the frame
    return self.frame

start(self)

Launches the internal Threaded Frames Extractor daemon

Returns: A reference to the ScreenGear class object.

Source code in vidgear/gears/screengear.py
def start(self):
    """
    Launches the internal *Threaded Frames Extractor* daemon

    **Returns:** A reference to the ScreenGear class object.
    """
    self.__thread = Thread(target=self.__update, name="ScreenGear", args=())
    self.__thread.daemon = True
    self.__thread.start()
    if self.__backend == "dxcam":
        self.__capture_object.start(
            target_fps=self.__target_fps,
            video_mode=True,
        )
        self.__logging and self.__target_fps and logger.debug(
            "Targeting FPS: {}".format(self.__target_fps)
        )
    return self

stop(self)

Safely terminates the thread, and release the resources.

Source code in vidgear/gears/screengear.py
def stop(self):
    """
    Safely terminates the thread, and release the resources.
    """
    self.__logging and logger.debug("Terminating ScreenGear Processes.")

    # indicate that the thread should be terminate
    self.__terminate.set()

    # wait until stream resources are released (producer thread might be still grabbing frame)
    not (self.__thread is None) and self.__thread.join()