Salta para o conteúdo

Referência para ultralytics/data/loaders.py

Nota

Este ficheiro está disponível em https://github.com/ultralytics/ ultralytics/blob/main/ ultralytics/data/loaders .py. Se detectares um problema, por favor ajuda a corrigi-lo contribuindo com um Pull Request 🛠️. Obrigado 🙏!



ultralytics.data.loaders.SourceTypes dataclass

Classe para representar vários tipos de fontes de entrada para previsões.

Código fonte em ultralytics/data/loaders.py
@dataclass
class SourceTypes:
    """Class to represent various types of input sources for predictions."""

    webcam: bool = False
    screenshot: bool = False
    from_img: bool = False
    tensor: bool = False



ultralytics.data.loaders.LoadStreams

Carrega vários tipos de fluxos de vídeo.

Adequado para utilização com yolo predict source='rtsp://example.com/media.mp4'O software de gestão de tráfego, suporta fluxos RTSP, RTMP, HTTP e TCP.

Atributos:

Nome Tipo Descrição
sources str

Os caminhos de entrada de origem ou URLs para os fluxos de vídeo.

vid_stride int

Stride da taxa de fotogramas do vídeo, predefinida para 1.

buffer bool

Se deves colocar fluxos de entrada em buffer, a predefinição é False.

running bool

Sinalizador para indicar se o thread de streaming está em execução.

mode str

Define "stream" para indicar a captura em tempo real.

imgs list

Lista de fotogramas de imagem para cada fluxo.

fps list

Lista de FPS para cada fluxo.

frames list

Lista o total de fotogramas para cada fluxo.

threads list

Lista de threads para cada fluxo.

shape list

Lista de formas para cada fluxo.

caps list

Lista de objectos cv2.VideoCapture para cada fluxo.

bs int

Tamanho do lote para processamento.

Métodos:

Nome Descrição
__init__

Inicializa o carregador de fluxo.

update

Lê quadros de fluxo no thread daemon.

close

Fecha o carregador de fluxo e liberta os recursos.

__iter__

Devolve um objeto iterador para a classe.

__next__

Devolve caminhos de origem, imagens transformadas e originais para processamento.

__len__

Devolve o comprimento do objeto fontes.

Código fonte em ultralytics/data/loaders.py
class LoadStreams:
    """
    Stream Loader for various types of video streams.

    Suitable for use with `yolo predict source='rtsp://example.com/media.mp4'`, supports RTSP, RTMP, HTTP, and TCP streams.

    Attributes:
        sources (str): The source input paths or URLs for the video streams.
        vid_stride (int): Video frame-rate stride, defaults to 1.
        buffer (bool): Whether to buffer input streams, defaults to False.
        running (bool): Flag to indicate if the streaming thread is running.
        mode (str): Set to 'stream' indicating real-time capture.
        imgs (list): List of image frames for each stream.
        fps (list): List of FPS for each stream.
        frames (list): List of total frames for each stream.
        threads (list): List of threads for each stream.
        shape (list): List of shapes for each stream.
        caps (list): List of cv2.VideoCapture objects for each stream.
        bs (int): Batch size for processing.

    Methods:
        __init__: Initialize the stream loader.
        update: Read stream frames in daemon thread.
        close: Close stream loader and release resources.
        __iter__: Returns an iterator object for the class.
        __next__: Returns source paths, transformed, and original images for processing.
        __len__: Return the length of the sources object.
    """

    def __init__(self, sources="file.streams", vid_stride=1, buffer=False):
        """Initialize instance variables and check for consistent input stream shapes."""
        torch.backends.cudnn.benchmark = True  # faster for fixed-size inference
        self.buffer = buffer  # buffer input streams
        self.running = True  # running flag for Thread
        self.mode = "stream"
        self.vid_stride = vid_stride  # video frame-rate stride

        sources = Path(sources).read_text().rsplit() if os.path.isfile(sources) else [sources]
        n = len(sources)
        self.fps = [0] * n  # frames per second
        self.frames = [0] * n
        self.threads = [None] * n
        self.caps = [None] * n  # video capture objects
        self.imgs = [[] for _ in range(n)]  # images
        self.shape = [[] for _ in range(n)]  # image shapes
        self.sources = [ops.clean_str(x) for x in sources]  # clean source names for later
        for i, s in enumerate(sources):  # index, source
            # Start thread to read frames from video stream
            st = f"{i + 1}/{n}: {s}... "
            if urlparse(s).hostname in ("www.youtube.com", "youtube.com", "youtu.be"):  # if source is YouTube video
                # YouTube format i.e. 'https://www.youtube.com/watch?v=Zgi9g1ksQHc' or 'https://youtu.be/LNwODJXcvt4'
                s = get_best_youtube_url(s)
            s = eval(s) if s.isnumeric() else s  # i.e. s = '0' local webcam
            if s == 0 and (is_colab() or is_kaggle()):
                raise NotImplementedError(
                    "'source=0' webcam not supported in Colab and Kaggle notebooks. "
                    "Try running 'source=0' in a local environment."
                )
            self.caps[i] = cv2.VideoCapture(s)  # store video capture object
            if not self.caps[i].isOpened():
                raise ConnectionError(f"{st}Failed to open {s}")
            w = int(self.caps[i].get(cv2.CAP_PROP_FRAME_WIDTH))
            h = int(self.caps[i].get(cv2.CAP_PROP_FRAME_HEIGHT))
            fps = self.caps[i].get(cv2.CAP_PROP_FPS)  # warning: may return 0 or nan
            self.frames[i] = max(int(self.caps[i].get(cv2.CAP_PROP_FRAME_COUNT)), 0) or float(
                "inf"
            )  # infinite stream fallback
            self.fps[i] = max((fps if math.isfinite(fps) else 0) % 100, 0) or 30  # 30 FPS fallback

            success, im = self.caps[i].read()  # guarantee first frame
            if not success or im is None:
                raise ConnectionError(f"{st}Failed to read images from {s}")
            self.imgs[i].append(im)
            self.shape[i] = im.shape
            self.threads[i] = Thread(target=self.update, args=([i, self.caps[i], s]), daemon=True)
            LOGGER.info(f"{st}Success ✅ ({self.frames[i]} frames of shape {w}x{h} at {self.fps[i]:.2f} FPS)")
            self.threads[i].start()
        LOGGER.info("")  # newline

        # Check for common shapes
        self.bs = self.__len__()

    def update(self, i, cap, stream):
        """Read stream `i` frames in daemon thread."""
        n, f = 0, self.frames[i]  # frame number, frame array
        while self.running and cap.isOpened() and n < (f - 1):
            if len(self.imgs[i]) < 30:  # keep a <=30-image buffer
                n += 1
                cap.grab()  # .read() = .grab() followed by .retrieve()
                if n % self.vid_stride == 0:
                    success, im = cap.retrieve()
                    if not success:
                        im = np.zeros(self.shape[i], dtype=np.uint8)
                        LOGGER.warning("WARNING ⚠️ Video stream unresponsive, please check your IP camera connection.")
                        cap.open(stream)  # re-open stream if signal was lost
                    if self.buffer:
                        self.imgs[i].append(im)
                    else:
                        self.imgs[i] = [im]
            else:
                time.sleep(0.01)  # wait until the buffer is empty

    def close(self):
        """Close stream loader and release resources."""
        self.running = False  # stop flag for Thread
        for thread in self.threads:
            if thread.is_alive():
                thread.join(timeout=5)  # Add timeout
        for cap in self.caps:  # Iterate through the stored VideoCapture objects
            try:
                cap.release()  # release video capture
            except Exception as e:
                LOGGER.warning(f"WARNING ⚠️ Could not release VideoCapture object: {e}")
        cv2.destroyAllWindows()

    def __iter__(self):
        """Iterates through YOLO image feed and re-opens unresponsive streams."""
        self.count = -1
        return self

    def __next__(self):
        """Returns source paths, transformed and original images for processing."""
        self.count += 1

        images = []
        for i, x in enumerate(self.imgs):
            # Wait until a frame is available in each buffer
            while not x:
                if not self.threads[i].is_alive() or cv2.waitKey(1) == ord("q"):  # q to quit
                    self.close()
                    raise StopIteration
                time.sleep(1 / min(self.fps))
                x = self.imgs[i]
                if not x:
                    LOGGER.warning(f"WARNING ⚠️ Waiting for stream {i}")

            # Get and remove the first frame from imgs buffer
            if self.buffer:
                images.append(x.pop(0))

            # Get the last frame, and clear the rest from the imgs buffer
            else:
                images.append(x.pop(-1) if x else np.zeros(self.shape[i], dtype=np.uint8))
                x.clear()

        return self.sources, images, None, ""

    def __len__(self):
        """Return the length of the sources object."""
        return len(self.sources)  # 1E12 frames = 32 streams at 30 FPS for 30 years

__init__(sources='file.streams', vid_stride=1, buffer=False)

Inicializa as variáveis de instância e verifica se as formas do fluxo de entrada são consistentes.

Código fonte em ultralytics/data/loaders.py
def __init__(self, sources="file.streams", vid_stride=1, buffer=False):
    """Initialize instance variables and check for consistent input stream shapes."""
    torch.backends.cudnn.benchmark = True  # faster for fixed-size inference
    self.buffer = buffer  # buffer input streams
    self.running = True  # running flag for Thread
    self.mode = "stream"
    self.vid_stride = vid_stride  # video frame-rate stride

    sources = Path(sources).read_text().rsplit() if os.path.isfile(sources) else [sources]
    n = len(sources)
    self.fps = [0] * n  # frames per second
    self.frames = [0] * n
    self.threads = [None] * n
    self.caps = [None] * n  # video capture objects
    self.imgs = [[] for _ in range(n)]  # images
    self.shape = [[] for _ in range(n)]  # image shapes
    self.sources = [ops.clean_str(x) for x in sources]  # clean source names for later
    for i, s in enumerate(sources):  # index, source
        # Start thread to read frames from video stream
        st = f"{i + 1}/{n}: {s}... "
        if urlparse(s).hostname in ("www.youtube.com", "youtube.com", "youtu.be"):  # if source is YouTube video
            # YouTube format i.e. 'https://www.youtube.com/watch?v=Zgi9g1ksQHc' or 'https://youtu.be/LNwODJXcvt4'
            s = get_best_youtube_url(s)
        s = eval(s) if s.isnumeric() else s  # i.e. s = '0' local webcam
        if s == 0 and (is_colab() or is_kaggle()):
            raise NotImplementedError(
                "'source=0' webcam not supported in Colab and Kaggle notebooks. "
                "Try running 'source=0' in a local environment."
            )
        self.caps[i] = cv2.VideoCapture(s)  # store video capture object
        if not self.caps[i].isOpened():
            raise ConnectionError(f"{st}Failed to open {s}")
        w = int(self.caps[i].get(cv2.CAP_PROP_FRAME_WIDTH))
        h = int(self.caps[i].get(cv2.CAP_PROP_FRAME_HEIGHT))
        fps = self.caps[i].get(cv2.CAP_PROP_FPS)  # warning: may return 0 or nan
        self.frames[i] = max(int(self.caps[i].get(cv2.CAP_PROP_FRAME_COUNT)), 0) or float(
            "inf"
        )  # infinite stream fallback
        self.fps[i] = max((fps if math.isfinite(fps) else 0) % 100, 0) or 30  # 30 FPS fallback

        success, im = self.caps[i].read()  # guarantee first frame
        if not success or im is None:
            raise ConnectionError(f"{st}Failed to read images from {s}")
        self.imgs[i].append(im)
        self.shape[i] = im.shape
        self.threads[i] = Thread(target=self.update, args=([i, self.caps[i], s]), daemon=True)
        LOGGER.info(f"{st}Success ✅ ({self.frames[i]} frames of shape {w}x{h} at {self.fps[i]:.2f} FPS)")
        self.threads[i].start()
    LOGGER.info("")  # newline

    # Check for common shapes
    self.bs = self.__len__()

__iter__()

Itera através do feed de imagens YOLO e reabre os fluxos que não respondem.

Código fonte em ultralytics/data/loaders.py
def __iter__(self):
    """Iterates through YOLO image feed and re-opens unresponsive streams."""
    self.count = -1
    return self

__len__()

Devolve o comprimento do objeto fontes.

Código fonte em ultralytics/data/loaders.py
def __len__(self):
    """Return the length of the sources object."""
    return len(self.sources)  # 1E12 frames = 32 streams at 30 FPS for 30 years

__next__()

Devolve caminhos de origem, imagens transformadas e originais para processamento.

Código fonte em ultralytics/data/loaders.py
def __next__(self):
    """Returns source paths, transformed and original images for processing."""
    self.count += 1

    images = []
    for i, x in enumerate(self.imgs):
        # Wait until a frame is available in each buffer
        while not x:
            if not self.threads[i].is_alive() or cv2.waitKey(1) == ord("q"):  # q to quit
                self.close()
                raise StopIteration
            time.sleep(1 / min(self.fps))
            x = self.imgs[i]
            if not x:
                LOGGER.warning(f"WARNING ⚠️ Waiting for stream {i}")

        # Get and remove the first frame from imgs buffer
        if self.buffer:
            images.append(x.pop(0))

        # Get the last frame, and clear the rest from the imgs buffer
        else:
            images.append(x.pop(-1) if x else np.zeros(self.shape[i], dtype=np.uint8))
            x.clear()

    return self.sources, images, None, ""

close()

Fecha o carregador de fluxo e liberta os recursos.

Código fonte em ultralytics/data/loaders.py
def close(self):
    """Close stream loader and release resources."""
    self.running = False  # stop flag for Thread
    for thread in self.threads:
        if thread.is_alive():
            thread.join(timeout=5)  # Add timeout
    for cap in self.caps:  # Iterate through the stored VideoCapture objects
        try:
            cap.release()  # release video capture
        except Exception as e:
            LOGGER.warning(f"WARNING ⚠️ Could not release VideoCapture object: {e}")
    cv2.destroyAllWindows()

update(i, cap, stream)

Lê o fluxo i quadros na thread daemon.

Código fonte em ultralytics/data/loaders.py
def update(self, i, cap, stream):
    """Read stream `i` frames in daemon thread."""
    n, f = 0, self.frames[i]  # frame number, frame array
    while self.running and cap.isOpened() and n < (f - 1):
        if len(self.imgs[i]) < 30:  # keep a <=30-image buffer
            n += 1
            cap.grab()  # .read() = .grab() followed by .retrieve()
            if n % self.vid_stride == 0:
                success, im = cap.retrieve()
                if not success:
                    im = np.zeros(self.shape[i], dtype=np.uint8)
                    LOGGER.warning("WARNING ⚠️ Video stream unresponsive, please check your IP camera connection.")
                    cap.open(stream)  # re-open stream if signal was lost
                if self.buffer:
                    self.imgs[i].append(im)
                else:
                    self.imgs[i] = [im]
        else:
            time.sleep(0.01)  # wait until the buffer is empty



ultralytics.data.loaders.LoadScreenshots

YOLOv8 carregador de dados de capturas de ecrã.

Esta classe gere o carregamento de imagens de capturas de ecrã para processamento com YOLOv8. Adequada para utilização com yolo predict source=screen.

Atributos:

Nome Tipo Descrição
source str

A entrada de origem que indica o ecrã a capturar.

screen int

O número do ecrã a capturar.

left int

A coordenada esquerda da área de captura de ecrã.

top int

A coordenada superior da área de captura de ecrã.

width int

A largura da área de captura de ecrã.

height int

A altura da área de captura de ecrã.

mode str

Define "stream" para indicar a captura em tempo real.

frame int

Contador de fotogramas capturados.

sct mss

Objeto de captura de ecrã de mss biblioteca.

bs int

Tamanho do lote, definido para 1.

monitor dict

Monitoriza os detalhes da configuração.

Métodos:

Nome Descrição
__iter__

Devolve um objeto iterador.

__next__

Captura a próxima captura de ecrã e devolve-a.

Código fonte em ultralytics/data/loaders.py
class LoadScreenshots:
    """
    YOLOv8 screenshot dataloader.

    This class manages the loading of screenshot images for processing with YOLOv8.
    Suitable for use with `yolo predict source=screen`.

    Attributes:
        source (str): The source input indicating which screen to capture.
        screen (int): The screen number to capture.
        left (int): The left coordinate for screen capture area.
        top (int): The top coordinate for screen capture area.
        width (int): The width of the screen capture area.
        height (int): The height of the screen capture area.
        mode (str): Set to 'stream' indicating real-time capture.
        frame (int): Counter for captured frames.
        sct (mss.mss): Screen capture object from `mss` library.
        bs (int): Batch size, set to 1.
        monitor (dict): Monitor configuration details.

    Methods:
        __iter__: Returns an iterator object.
        __next__: Captures the next screenshot and returns it.
    """

    def __init__(self, source):
        """Source = [screen_number left top width height] (pixels)."""
        check_requirements("mss")
        import mss  # noqa

        source, *params = source.split()
        self.screen, left, top, width, height = 0, None, None, None, None  # default to full screen 0
        if len(params) == 1:
            self.screen = int(params[0])
        elif len(params) == 4:
            left, top, width, height = (int(x) for x in params)
        elif len(params) == 5:
            self.screen, left, top, width, height = (int(x) for x in params)
        self.mode = "stream"
        self.frame = 0
        self.sct = mss.mss()
        self.bs = 1

        # Parse monitor shape
        monitor = self.sct.monitors[self.screen]
        self.top = monitor["top"] if top is None else (monitor["top"] + top)
        self.left = monitor["left"] if left is None else (monitor["left"] + left)
        self.width = width or monitor["width"]
        self.height = height or monitor["height"]
        self.monitor = {"left": self.left, "top": self.top, "width": self.width, "height": self.height}

    def __iter__(self):
        """Returns an iterator of the object."""
        return self

    def __next__(self):
        """mss screen capture: get raw pixels from the screen as np array."""
        im0 = np.asarray(self.sct.grab(self.monitor))[:, :, :3]  # BGRA to BGR
        s = f"screen {self.screen} (LTWH): {self.left},{self.top},{self.width},{self.height}: "

        self.frame += 1
        return [str(self.screen)], [im0], None, s  # screen, img, vid_cap, string

__init__(source)

Fonte = [número_do_ecrã esquerda topo largura altura] (pixels).

Código fonte em ultralytics/data/loaders.py
def __init__(self, source):
    """Source = [screen_number left top width height] (pixels)."""
    check_requirements("mss")
    import mss  # noqa

    source, *params = source.split()
    self.screen, left, top, width, height = 0, None, None, None, None  # default to full screen 0
    if len(params) == 1:
        self.screen = int(params[0])
    elif len(params) == 4:
        left, top, width, height = (int(x) for x in params)
    elif len(params) == 5:
        self.screen, left, top, width, height = (int(x) for x in params)
    self.mode = "stream"
    self.frame = 0
    self.sct = mss.mss()
    self.bs = 1

    # Parse monitor shape
    monitor = self.sct.monitors[self.screen]
    self.top = monitor["top"] if top is None else (monitor["top"] + top)
    self.left = monitor["left"] if left is None else (monitor["left"] + left)
    self.width = width or monitor["width"]
    self.height = height or monitor["height"]
    self.monitor = {"left": self.left, "top": self.top, "width": self.width, "height": self.height}

__iter__()

Devolve um iterador do objeto.

Código fonte em ultralytics/data/loaders.py
def __iter__(self):
    """Returns an iterator of the object."""
    return self

__next__()

mss screen capture: obtém pixels brutos do ecrã como uma matriz np.

Código fonte em ultralytics/data/loaders.py
def __next__(self):
    """mss screen capture: get raw pixels from the screen as np array."""
    im0 = np.asarray(self.sct.grab(self.monitor))[:, :, :3]  # BGRA to BGR
    s = f"screen {self.screen} (LTWH): {self.left},{self.top},{self.width},{self.height}: "

    self.frame += 1
    return [str(self.screen)], [im0], None, s  # screen, img, vid_cap, string



ultralytics.data.loaders.LoadImages

YOLOv8 imagem/vídeo dataloader.

Esta classe gere o carregamento e o pré-processamento de dados de imagem e vídeo para YOLOv8. Suporta o carregamento a partir de vários formatos, incluindo ficheiros de imagem simples, ficheiros de vídeo e listas de caminhos de imagem e vídeo.

Atributos:

Nome Tipo Descrição
files list

Lista de caminhos de ficheiros de imagem e vídeo.

nf int

Número total de ficheiros (imagens e vídeos).

video_flag list

Sinalizadores que indicam se um ficheiro é um vídeo (Verdadeiro) ou uma imagem (Falso).

mode str

Modo atual, "imagem" ou "vídeo".

vid_stride int

Desloca-se para a taxa de fotogramas do vídeo; a predefinição é 1.

bs int

Tamanho do lote, definido como 1 para esta classe.

cap VideoCapture

Objeto de captura de vídeo para OpenCV.

frame int

Contador de fotogramas para vídeo.

frames int

Número total de fotogramas no vídeo.

count int

Contador para iteração, inicializado a 0 durante __iter__().

Métodos:

Nome Descrição
_new_video

Cria um novo objeto cv2.VideoCapture para um determinado caminho de vídeo.

Código fonte em ultralytics/data/loaders.py
class LoadImages:
    """
    YOLOv8 image/video dataloader.

    This class manages the loading and pre-processing of image and video data for YOLOv8. It supports loading from
    various formats, including single image files, video files, and lists of image and video paths.

    Attributes:
        files (list): List of image and video file paths.
        nf (int): Total number of files (images and videos).
        video_flag (list): Flags indicating whether a file is a video (True) or an image (False).
        mode (str): Current mode, 'image' or 'video'.
        vid_stride (int): Stride for video frame-rate, defaults to 1.
        bs (int): Batch size, set to 1 for this class.
        cap (cv2.VideoCapture): Video capture object for OpenCV.
        frame (int): Frame counter for video.
        frames (int): Total number of frames in the video.
        count (int): Counter for iteration, initialized at 0 during `__iter__()`.

    Methods:
        _new_video(path): Create a new cv2.VideoCapture object for a given video path.
    """

    def __init__(self, path, vid_stride=1):
        """Initialize the Dataloader and raise FileNotFoundError if file not found."""
        parent = None
        if isinstance(path, str) and Path(path).suffix == ".txt":  # *.txt file with img/vid/dir on each line
            parent = Path(path).parent
            path = Path(path).read_text().splitlines()  # list of sources
        files = []
        for p in sorted(path) if isinstance(path, (list, tuple)) else [path]:
            a = str(Path(p).absolute())  # do not use .resolve() https://github.com/ultralytics/ultralytics/issues/2912
            if "*" in a:
                files.extend(sorted(glob.glob(a, recursive=True)))  # glob
            elif os.path.isdir(a):
                files.extend(sorted(glob.glob(os.path.join(a, "*.*"))))  # dir
            elif os.path.isfile(a):
                files.append(a)  # files (absolute or relative to CWD)
            elif parent and (parent / p).is_file():
                files.append(str((parent / p).absolute()))  # files (relative to *.txt file parent)
            else:
                raise FileNotFoundError(f"{p} does not exist")

        images = [x for x in files if x.split(".")[-1].lower() in IMG_FORMATS]
        videos = [x for x in files if x.split(".")[-1].lower() in VID_FORMATS]
        ni, nv = len(images), len(videos)

        self.files = images + videos
        self.nf = ni + nv  # number of files
        self.video_flag = [False] * ni + [True] * nv
        self.mode = "image"
        self.vid_stride = vid_stride  # video frame-rate stride
        self.bs = 1
        if any(videos):
            self._new_video(videos[0])  # new video
        else:
            self.cap = None
        if self.nf == 0:
            raise FileNotFoundError(
                f"No images or videos found in {p}. "
                f"Supported formats are:\nimages: {IMG_FORMATS}\nvideos: {VID_FORMATS}"
            )

    def __iter__(self):
        """Returns an iterator object for VideoStream or ImageFolder."""
        self.count = 0
        return self

    def __next__(self):
        """Return next image, path and metadata from dataset."""
        if self.count == self.nf:
            raise StopIteration
        path = self.files[self.count]

        if self.video_flag[self.count]:
            # Read video
            self.mode = "video"
            for _ in range(self.vid_stride):
                self.cap.grab()
            success, im0 = self.cap.retrieve()
            while not success:
                self.count += 1
                self.cap.release()
                if self.count == self.nf:  # last video
                    raise StopIteration
                path = self.files[self.count]
                self._new_video(path)
                success, im0 = self.cap.read()

            self.frame += 1
            # im0 = self._cv2_rotate(im0)  # for use if cv2 autorotation is False
            s = f"video {self.count + 1}/{self.nf} ({self.frame}/{self.frames}) {path}: "

        else:
            # Read image
            self.count += 1
            im0 = cv2.imread(path)  # BGR
            if im0 is None:
                raise FileNotFoundError(f"Image Not Found {path}")
            s = f"image {self.count}/{self.nf} {path}: "

        return [path], [im0], self.cap, s

    def _new_video(self, path):
        """Create a new video capture object."""
        self.frame = 0
        self.cap = cv2.VideoCapture(path)
        self.frames = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT) / self.vid_stride)

    def __len__(self):
        """Returns the number of files in the object."""
        return self.nf  # number of files

__init__(path, vid_stride=1)

Inicializa o Dataloader e levanta FileNotFoundError se o ficheiro não for encontrado.

Código fonte em ultralytics/data/loaders.py
def __init__(self, path, vid_stride=1):
    """Initialize the Dataloader and raise FileNotFoundError if file not found."""
    parent = None
    if isinstance(path, str) and Path(path).suffix == ".txt":  # *.txt file with img/vid/dir on each line
        parent = Path(path).parent
        path = Path(path).read_text().splitlines()  # list of sources
    files = []
    for p in sorted(path) if isinstance(path, (list, tuple)) else [path]:
        a = str(Path(p).absolute())  # do not use .resolve() https://github.com/ultralytics/ultralytics/issues/2912
        if "*" in a:
            files.extend(sorted(glob.glob(a, recursive=True)))  # glob
        elif os.path.isdir(a):
            files.extend(sorted(glob.glob(os.path.join(a, "*.*"))))  # dir
        elif os.path.isfile(a):
            files.append(a)  # files (absolute or relative to CWD)
        elif parent and (parent / p).is_file():
            files.append(str((parent / p).absolute()))  # files (relative to *.txt file parent)
        else:
            raise FileNotFoundError(f"{p} does not exist")

    images = [x for x in files if x.split(".")[-1].lower() in IMG_FORMATS]
    videos = [x for x in files if x.split(".")[-1].lower() in VID_FORMATS]
    ni, nv = len(images), len(videos)

    self.files = images + videos
    self.nf = ni + nv  # number of files
    self.video_flag = [False] * ni + [True] * nv
    self.mode = "image"
    self.vid_stride = vid_stride  # video frame-rate stride
    self.bs = 1
    if any(videos):
        self._new_video(videos[0])  # new video
    else:
        self.cap = None
    if self.nf == 0:
        raise FileNotFoundError(
            f"No images or videos found in {p}. "
            f"Supported formats are:\nimages: {IMG_FORMATS}\nvideos: {VID_FORMATS}"
        )

__iter__()

Devolve um objeto iterador para VideoStream ou ImageFolder.

Código fonte em ultralytics/data/loaders.py
def __iter__(self):
    """Returns an iterator object for VideoStream or ImageFolder."""
    self.count = 0
    return self

__len__()

Devolve o número de ficheiros no objeto.

Código fonte em ultralytics/data/loaders.py
def __len__(self):
    """Returns the number of files in the object."""
    return self.nf  # number of files

__next__()

Devolve a imagem seguinte, o caminho e os metadados do conjunto de dados.

Código fonte em ultralytics/data/loaders.py
def __next__(self):
    """Return next image, path and metadata from dataset."""
    if self.count == self.nf:
        raise StopIteration
    path = self.files[self.count]

    if self.video_flag[self.count]:
        # Read video
        self.mode = "video"
        for _ in range(self.vid_stride):
            self.cap.grab()
        success, im0 = self.cap.retrieve()
        while not success:
            self.count += 1
            self.cap.release()
            if self.count == self.nf:  # last video
                raise StopIteration
            path = self.files[self.count]
            self._new_video(path)
            success, im0 = self.cap.read()

        self.frame += 1
        # im0 = self._cv2_rotate(im0)  # for use if cv2 autorotation is False
        s = f"video {self.count + 1}/{self.nf} ({self.frame}/{self.frames}) {path}: "

    else:
        # Read image
        self.count += 1
        im0 = cv2.imread(path)  # BGR
        if im0 is None:
            raise FileNotFoundError(f"Image Not Found {path}")
        s = f"image {self.count}/{self.nf} {path}: "

    return [path], [im0], self.cap, s



ultralytics.data.loaders.LoadPilAndNumpy

Carrega imagens de matrizes PIL e Numpy para processamento em lote.

Esta classe foi concebida para gerir o carregamento e o pré-processamento de dados de imagem dos formatos PIL e Numpy. Executa a validação básica e a conversão de formatos para garantir que as imagens estão no formato necessário para o processamento a jusante.

Atributos:

Nome Tipo Descrição
paths list

Lista de caminhos de imagem ou nomes de ficheiros gerados automaticamente.

im0 list

Lista de imagens armazenadas como arrays Numpy.

mode str

Tipo de dados que está a ser processado; a predefinição é "imagem".

bs int

Tamanho do lote, equivalente ao comprimento de im0.

count int

Contador para iteração, inicializado a 0 durante __iter__().

Métodos:

Nome Descrição
_single_check

Valida e formata uma única imagem para uma matriz Numpy.

Código fonte em ultralytics/data/loaders.py
class LoadPilAndNumpy:
    """
    Load images from PIL and Numpy arrays for batch processing.

    This class is designed to manage loading and pre-processing of image data from both PIL and Numpy formats.
    It performs basic validation and format conversion to ensure that the images are in the required format for
    downstream processing.

    Attributes:
        paths (list): List of image paths or autogenerated filenames.
        im0 (list): List of images stored as Numpy arrays.
        mode (str): Type of data being processed, defaults to 'image'.
        bs (int): Batch size, equivalent to the length of `im0`.
        count (int): Counter for iteration, initialized at 0 during `__iter__()`.

    Methods:
        _single_check(im): Validate and format a single image to a Numpy array.
    """

    def __init__(self, im0):
        """Initialize PIL and Numpy Dataloader."""
        if not isinstance(im0, list):
            im0 = [im0]
        self.paths = [getattr(im, "filename", f"image{i}.jpg") for i, im in enumerate(im0)]
        self.im0 = [self._single_check(im) for im in im0]
        self.mode = "image"
        # Generate fake paths
        self.bs = len(self.im0)

    @staticmethod
    def _single_check(im):
        """Validate and format an image to numpy array."""
        assert isinstance(im, (Image.Image, np.ndarray)), f"Expected PIL/np.ndarray image type, but got {type(im)}"
        if isinstance(im, Image.Image):
            if im.mode != "RGB":
                im = im.convert("RGB")
            im = np.asarray(im)[:, :, ::-1]
            im = np.ascontiguousarray(im)  # contiguous
        return im

    def __len__(self):
        """Returns the length of the 'im0' attribute."""
        return len(self.im0)

    def __next__(self):
        """Returns batch paths, images, processed images, None, ''."""
        if self.count == 1:  # loop only once as it's batch inference
            raise StopIteration
        self.count += 1
        return self.paths, self.im0, None, ""

    def __iter__(self):
        """Enables iteration for class LoadPilAndNumpy."""
        self.count = 0
        return self

__init__(im0)

Inicializa o PIL e o Numpy Dataloader.

Código fonte em ultralytics/data/loaders.py
def __init__(self, im0):
    """Initialize PIL and Numpy Dataloader."""
    if not isinstance(im0, list):
        im0 = [im0]
    self.paths = [getattr(im, "filename", f"image{i}.jpg") for i, im in enumerate(im0)]
    self.im0 = [self._single_check(im) for im in im0]
    self.mode = "image"
    # Generate fake paths
    self.bs = len(self.im0)

__iter__()

Ativa a iteração para a classe LoadPilAndNumpy.

Código fonte em ultralytics/data/loaders.py
def __iter__(self):
    """Enables iteration for class LoadPilAndNumpy."""
    self.count = 0
    return self

__len__()

Devolve o comprimento do atributo 'im0'.

Código fonte em ultralytics/data/loaders.py
def __len__(self):
    """Returns the length of the 'im0' attribute."""
    return len(self.im0)

__next__()

Devolve caminhos de lote, imagens, imagens processadas, None, ''.

Código fonte em ultralytics/data/loaders.py
def __next__(self):
    """Returns batch paths, images, processed images, None, ''."""
    if self.count == 1:  # loop only once as it's batch inference
        raise StopIteration
    self.count += 1
    return self.paths, self.im0, None, ""



ultralytics.data.loaders.LoadTensor

Carrega imagens de torch.Tensor dados.

Esta classe gere o carregamento e o pré-processamento de dados de imagem de PyTorch tensores para processamento posterior.

Atributos:

Nome Tipo Descrição
im0 Tensor

A entrada tensor que contém a(s) imagem(ns).

bs int

Tamanho do lote, inferido a partir da forma de im0.

mode str

Modo atual, definido como "imagem".

paths list

Lista de caminhos de imagem ou nomes de ficheiros.

count int

Contador para iteração, inicializado a 0 durante __iter__().

Métodos:

Nome Descrição
_single_check

Valida e possivelmente modifica a entrada tensor.

Código fonte em ultralytics/data/loaders.py
class LoadTensor:
    """
    Load images from torch.Tensor data.

    This class manages the loading and pre-processing of image data from PyTorch tensors for further processing.

    Attributes:
        im0 (torch.Tensor): The input tensor containing the image(s).
        bs (int): Batch size, inferred from the shape of `im0`.
        mode (str): Current mode, set to 'image'.
        paths (list): List of image paths or filenames.
        count (int): Counter for iteration, initialized at 0 during `__iter__()`.

    Methods:
        _single_check(im, stride): Validate and possibly modify the input tensor.
    """

    def __init__(self, im0) -> None:
        """Initialize Tensor Dataloader."""
        self.im0 = self._single_check(im0)
        self.bs = self.im0.shape[0]
        self.mode = "image"
        self.paths = [getattr(im, "filename", f"image{i}.jpg") for i, im in enumerate(im0)]

    @staticmethod
    def _single_check(im, stride=32):
        """Validate and format an image to torch.Tensor."""
        s = (
            f"WARNING ⚠️ torch.Tensor inputs should be BCHW i.e. shape(1, 3, 640, 640) "
            f"divisible by stride {stride}. Input shape{tuple(im.shape)} is incompatible."
        )
        if len(im.shape) != 4:
            if len(im.shape) != 3:
                raise ValueError(s)
            LOGGER.warning(s)
            im = im.unsqueeze(0)
        if im.shape[2] % stride or im.shape[3] % stride:
            raise ValueError(s)
        if im.max() > 1.0 + torch.finfo(im.dtype).eps:  # torch.float32 eps is 1.2e-07
            LOGGER.warning(
                f"WARNING ⚠️ torch.Tensor inputs should be normalized 0.0-1.0 but max value is {im.max()}. "
                f"Dividing input by 255."
            )
            im = im.float() / 255.0

        return im

    def __iter__(self):
        """Returns an iterator object."""
        self.count = 0
        return self

    def __next__(self):
        """Return next item in the iterator."""
        if self.count == 1:
            raise StopIteration
        self.count += 1
        return self.paths, self.im0, None, ""

    def __len__(self):
        """Returns the batch size."""
        return self.bs

__init__(im0)

Inicializa o Tensor Dataloader.

Código fonte em ultralytics/data/loaders.py
def __init__(self, im0) -> None:
    """Initialize Tensor Dataloader."""
    self.im0 = self._single_check(im0)
    self.bs = self.im0.shape[0]
    self.mode = "image"
    self.paths = [getattr(im, "filename", f"image{i}.jpg") for i, im in enumerate(im0)]

__iter__()

Devolve um objeto iterador.

Código fonte em ultralytics/data/loaders.py
def __iter__(self):
    """Returns an iterator object."""
    self.count = 0
    return self

__len__()

Devolve o tamanho do lote.

Código fonte em ultralytics/data/loaders.py
def __len__(self):
    """Returns the batch size."""
    return self.bs

__next__()

Devolve o item seguinte no iterador.

Código fonte em ultralytics/data/loaders.py
def __next__(self):
    """Return next item in the iterator."""
    if self.count == 1:
        raise StopIteration
    self.count += 1
    return self.paths, self.im0, None, ""



ultralytics.data.loaders.autocast_list(source)

Mescla uma lista de fontes de diferentes tipos em uma lista de matrizes numpy ou imagens PIL.

Código fonte em ultralytics/data/loaders.py
def autocast_list(source):
    """Merges a list of source of different types into a list of numpy arrays or PIL images."""
    files = []
    for im in source:
        if isinstance(im, (str, Path)):  # filename or uri
            files.append(Image.open(requests.get(im, stream=True).raw if str(im).startswith("http") else im))
        elif isinstance(im, (Image.Image, np.ndarray)):  # PIL or np Image
            files.append(im)
        else:
            raise TypeError(
                f"type {type(im).__name__} is not a supported Ultralytics prediction source type. \n"
                f"See https://docs.ultralytics.com/modes/predict for supported source types."
            )

    return files



ultralytics.data.loaders.get_best_youtube_url(url, use_pafy=True)

Recupera o URL do fluxo de vídeo MP4 de melhor qualidade de um determinado vídeo do YouTube.

Esta função utiliza a biblioteca pafy ou yt_dlp para extrair a informação do vídeo do YouTube. Em seguida, encontra o formato MP4 de maior encontra o formato MP4 de maior qualidade que tem codec de vídeo mas não tem codec de áudio, e devolve o URL deste fluxo de vídeo.

Parâmetros:

Nome Tipo Descrição Predefinição
url str

O URL do vídeo do YouTube.

necessário
use_pafy bool

Utiliza o pacote pafy, predefinição=True, caso contrário utiliza o pacote yt_dlp.

True

Devolve:

Tipo Descrição
str

O URL do fluxo de vídeo MP4 de melhor qualidade, ou Nenhum se não for encontrado um fluxo adequado.

Código fonte em ultralytics/data/loaders.py
def get_best_youtube_url(url, use_pafy=True):
    """
    Retrieves the URL of the best quality MP4 video stream from a given YouTube video.

    This function uses the pafy or yt_dlp library to extract the video info from YouTube. It then finds the highest
    quality MP4 format that has video codec but no audio codec, and returns the URL of this video stream.

    Args:
        url (str): The URL of the YouTube video.
        use_pafy (bool): Use the pafy package, default=True, otherwise use yt_dlp package.

    Returns:
        (str): The URL of the best quality MP4 video stream, or None if no suitable stream is found.
    """
    if use_pafy:
        check_requirements(("pafy", "youtube_dl==2020.12.2"))
        import pafy  # noqa

        return pafy.new(url).getbestvideo(preftype="mp4").url
    else:
        check_requirements("yt-dlp")
        import yt_dlp

        with yt_dlp.YoutubeDL({"quiet": True}) as ydl:
            info_dict = ydl.extract_info(url, download=False)  # extract info
        for f in reversed(info_dict.get("formats", [])):  # reversed because best is usually last
            # Find a format with video codec, no audio, *.mp4 extension at least 1920x1080 size
            good_size = (f.get("width") or 0) >= 1920 or (f.get("height") or 0) >= 1080
            if good_size and f["vcodec"] != "none" and f["acodec"] == "none" and f["ext"] == "mp4":
                return f.get("url")





Criado em 2023-11-12, Atualizado em 2023-11-25
Autores: glenn-jocher (3), Laughing-q (1)