Skip to content

Reference for ultralytics/data/base.py

Note

This file is available at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/data/base.py. If you spot a problem please help fix it by contributing a Pull Request ūüõ†ÔłŹ. Thank you ūüôŹ!


ultralytics.data.base.BaseDataset

BaseDataset(img_path, imgsz=640, cache=False, augment=True, hyp=DEFAULT_CFG, prefix='', rect=False, batch_size=16, stride=32, pad=0.5, single_cls=False, classes=None, fraction=1.0)

Bases: Dataset

Base dataset class for loading and processing image data.

Parameters:

Name Type Description Default
img_path str

Path to the folder containing images.

required
imgsz int

Image size. Defaults to 640.

640
cache bool

Cache images to RAM or disk during training. Defaults to False.

False
augment bool

If True, data augmentation is applied. Defaults to True.

True
hyp dict

Hyperparameters to apply data augmentation. Defaults to None.

DEFAULT_CFG
prefix str

Prefix to print in log messages. Defaults to ''.

''
rect bool

If True, rectangular training is used. Defaults to False.

False
batch_size int

Size of batches. Defaults to None.

16
stride int

Stride. Defaults to 32.

32
pad float

Padding. Defaults to 0.0.

0.5
single_cls bool

If True, single class training is used. Defaults to False.

False
classes list

List of included classes. Default is None.

None
fraction float

Fraction of dataset to utilize. Default is 1.0 (use all data).

1.0

Attributes:

Name Type Description
im_files list

List of image file paths.

labels list

List of label data dictionaries.

ni int

Number of images in the dataset.

ims list

List of loaded images.

npy_files list

List of numpy file paths.

transforms callable

Image transformation function.

Source code in ultralytics/data/base.py
def __init__(
    self,
    img_path,
    imgsz=640,
    cache=False,
    augment=True,
    hyp=DEFAULT_CFG,
    prefix="",
    rect=False,
    batch_size=16,
    stride=32,
    pad=0.5,
    single_cls=False,
    classes=None,
    fraction=1.0,
):
    """Initialize BaseDataset with given configuration and options."""
    super().__init__()
    self.img_path = img_path
    self.imgsz = imgsz
    self.augment = augment
    self.single_cls = single_cls
    self.prefix = prefix
    self.fraction = fraction
    self.im_files = self.get_img_files(self.img_path)
    self.labels = self.get_labels()
    self.update_labels(include_class=classes)  # single_cls and include_class
    self.ni = len(self.labels)  # number of images
    self.rect = rect
    self.batch_size = batch_size
    self.stride = stride
    self.pad = pad
    if self.rect:
        assert self.batch_size is not None
        self.set_rectangle()

    # Buffer thread for mosaic images
    self.buffer = []  # buffer size = batch size
    self.max_buffer_length = min((self.ni, self.batch_size * 8, 1000)) if self.augment else 0

    # Cache images (options are cache = True, False, None, "ram", "disk")
    self.ims, self.im_hw0, self.im_hw = [None] * self.ni, [None] * self.ni, [None] * self.ni
    self.npy_files = [Path(f).with_suffix(".npy") for f in self.im_files]
    self.cache = cache.lower() if isinstance(cache, str) else "ram" if cache is True else None
    if (self.cache == "ram" and self.check_cache_ram()) or self.cache == "disk":
        self.cache_images()

    # Transforms
    self.transforms = self.build_transforms(hyp=hyp)

__getitem__

__getitem__(index)

Returns transformed label information for given index.

Source code in ultralytics/data/base.py
def __getitem__(self, index):
    """Returns transformed label information for given index."""
    return self.transforms(self.get_image_and_label(index))

__len__

__len__()

Returns the length of the labels list for the dataset.

Source code in ultralytics/data/base.py
def __len__(self):
    """Returns the length of the labels list for the dataset."""
    return len(self.labels)

build_transforms

build_transforms(hyp=None)

Users can customize augmentations here.

Example
if self.augment:
    # Training transforms
    return Compose([])
else:
    # Val transforms
    return Compose([])
Source code in ultralytics/data/base.py
def build_transforms(self, hyp=None):
    """
    Users can customize augmentations here.

    Example:
        ```python
        if self.augment:
            # Training transforms
            return Compose([])
        else:
            # Val transforms
            return Compose([])
        ```
    """
    raise NotImplementedError

cache_images

cache_images()

Cache images to memory or disk.

Source code in ultralytics/data/base.py
def cache_images(self):
    """Cache images to memory or disk."""
    b, gb = 0, 1 << 30  # bytes of cached images, bytes per gigabytes
    fcn, storage = (self.cache_images_to_disk, "Disk") if self.cache == "disk" else (self.load_image, "RAM")
    with ThreadPool(NUM_THREADS) as pool:
        results = pool.imap(fcn, range(self.ni))
        pbar = TQDM(enumerate(results), total=self.ni, disable=LOCAL_RANK > 0)
        for i, x in pbar:
            if self.cache == "disk":
                b += self.npy_files[i].stat().st_size
            else:  # 'ram'
                self.ims[i], self.im_hw0[i], self.im_hw[i] = x  # im, hw_orig, hw_resized = load_image(self, i)
                b += self.ims[i].nbytes
            pbar.desc = f"{self.prefix}Caching images ({b / gb:.1f}GB {storage})"
        pbar.close()

cache_images_to_disk

cache_images_to_disk(i)

Saves an image as an *.npy file for faster loading.

Source code in ultralytics/data/base.py
def cache_images_to_disk(self, i):
    """Saves an image as an *.npy file for faster loading."""
    f = self.npy_files[i]
    if not f.exists():
        np.save(f.as_posix(), cv2.imread(self.im_files[i]), allow_pickle=False)

check_cache_ram

check_cache_ram(safety_margin=0.5)

Check image caching requirements vs available memory.

Source code in ultralytics/data/base.py
def check_cache_ram(self, safety_margin=0.5):
    """Check image caching requirements vs available memory."""
    b, gb = 0, 1 << 30  # bytes of cached images, bytes per gigabytes
    n = min(self.ni, 30)  # extrapolate from 30 random images
    for _ in range(n):
        im = cv2.imread(random.choice(self.im_files))  # sample image
        ratio = self.imgsz / max(im.shape[0], im.shape[1])  # max(h, w)  # ratio
        b += im.nbytes * ratio**2
    mem_required = b * self.ni / n * (1 + safety_margin)  # GB required to cache dataset into RAM
    mem = psutil.virtual_memory()
    success = mem_required < mem.available  # to cache or not to cache, that is the question
    if not success:
        self.cache = None
        LOGGER.info(
            f"{self.prefix}{mem_required / gb:.1f}GB RAM required to cache images "
            f"with {int(safety_margin * 100)}% safety margin but only "
            f"{mem.available / gb:.1f}/{mem.total / gb:.1f}GB available, not caching images ‚ö†ÔłŹ"
        )
    return success

get_image_and_label

get_image_and_label(index)

Get and return label information from the dataset.

Source code in ultralytics/data/base.py
def get_image_and_label(self, index):
    """Get and return label information from the dataset."""
    label = deepcopy(self.labels[index])  # requires deepcopy() https://github.com/ultralytics/ultralytics/pull/1948
    label.pop("shape", None)  # shape is for rect, remove it
    label["img"], label["ori_shape"], label["resized_shape"] = self.load_image(index)
    label["ratio_pad"] = (
        label["resized_shape"][0] / label["ori_shape"][0],
        label["resized_shape"][1] / label["ori_shape"][1],
    )  # for evaluation
    if self.rect:
        label["rect_shape"] = self.batch_shapes[self.batch[index]]
    return self.update_labels_info(label)

get_img_files

get_img_files(img_path)

Read image files.

Source code in ultralytics/data/base.py
def get_img_files(self, img_path):
    """Read image files."""
    try:
        f = []  # image files
        for p in img_path if isinstance(img_path, list) else [img_path]:
            p = Path(p)  # os-agnostic
            if p.is_dir():  # dir
                f += glob.glob(str(p / "**" / "*.*"), recursive=True)
                # F = list(p.rglob('*.*'))  # pathlib
            elif p.is_file():  # file
                with open(p) as t:
                    t = t.read().strip().splitlines()
                    parent = str(p.parent) + os.sep
                    f += [x.replace("./", parent) if x.startswith("./") else x for x in t]  # local to global path
                    # F += [p.parent / x.lstrip(os.sep) for x in t]  # local to global path (pathlib)
            else:
                raise FileNotFoundError(f"{self.prefix}{p} does not exist")
        im_files = sorted(x.replace("/", os.sep) for x in f if x.split(".")[-1].lower() in IMG_FORMATS)
        # self.img_files = sorted([x for x in f if x.suffix[1:].lower() in IMG_FORMATS])  # pathlib
        assert im_files, f"{self.prefix}No images found in {img_path}. {FORMATS_HELP_MSG}"
    except Exception as e:
        raise FileNotFoundError(f"{self.prefix}Error loading data from {img_path}\n{HELP_URL}") from e
    if self.fraction < 1:
        im_files = im_files[: round(len(im_files) * self.fraction)]  # retain a fraction of the dataset
    return im_files

get_labels

get_labels()

Users can customize their own format here.

Note

Ensure output is a dictionary with the following keys:

dict(
    im_file=im_file,
    shape=shape,  # format: (height, width)
    cls=cls,
    bboxes=bboxes, # xywh
    segments=segments,  # xy
    keypoints=keypoints, # xy
    normalized=True, # or False
    bbox_format="xyxy",  # or xywh, ltwh
)

Source code in ultralytics/data/base.py
def get_labels(self):
    """
    Users can customize their own format here.

    Note:
        Ensure output is a dictionary with the following keys:
        ```python
        dict(
            im_file=im_file,
            shape=shape,  # format: (height, width)
            cls=cls,
            bboxes=bboxes, # xywh
            segments=segments,  # xy
            keypoints=keypoints, # xy
            normalized=True, # or False
            bbox_format="xyxy",  # or xywh, ltwh
        )
        ```
    """
    raise NotImplementedError

load_image

load_image(i, rect_mode=True)

Loads 1 image from dataset index 'i', returns (im, resized hw).

Source code in ultralytics/data/base.py
def load_image(self, i, rect_mode=True):
    """Loads 1 image from dataset index 'i', returns (im, resized hw)."""
    im, f, fn = self.ims[i], self.im_files[i], self.npy_files[i]
    if im is None:  # not cached in RAM
        if fn.exists():  # load npy
            try:
                im = np.load(fn)
            except Exception as e:
                LOGGER.warning(f"{self.prefix}WARNING ‚ö†ÔłŹ Removing corrupt *.npy image file {fn} due to: {e}")
                Path(fn).unlink(missing_ok=True)
                im = cv2.imread(f)  # BGR
        else:  # read image
            im = cv2.imread(f)  # BGR
        if im is None:
            raise FileNotFoundError(f"Image Not Found {f}")

        h0, w0 = im.shape[:2]  # orig hw
        if rect_mode:  # resize long side to imgsz while maintaining aspect ratio
            r = self.imgsz / max(h0, w0)  # ratio
            if r != 1:  # if sizes are not equal
                w, h = (min(math.ceil(w0 * r), self.imgsz), min(math.ceil(h0 * r), self.imgsz))
                im = cv2.resize(im, (w, h), interpolation=cv2.INTER_LINEAR)
        elif not (h0 == w0 == self.imgsz):  # resize by stretching image to square imgsz
            im = cv2.resize(im, (self.imgsz, self.imgsz), interpolation=cv2.INTER_LINEAR)

        # Add to buffer if training with augmentations
        if self.augment:
            self.ims[i], self.im_hw0[i], self.im_hw[i] = im, (h0, w0), im.shape[:2]  # im, hw_original, hw_resized
            self.buffer.append(i)
            if 1 < len(self.buffer) >= self.max_buffer_length:  # prevent empty buffer
                j = self.buffer.pop(0)
                if self.cache != "ram":
                    self.ims[j], self.im_hw0[j], self.im_hw[j] = None, None, None

        return im, (h0, w0), im.shape[:2]

    return self.ims[i], self.im_hw0[i], self.im_hw[i]

set_rectangle

set_rectangle()

Sets the shape of bounding boxes for YOLO detections as rectangles.

Source code in ultralytics/data/base.py
def set_rectangle(self):
    """Sets the shape of bounding boxes for YOLO detections as rectangles."""
    bi = np.floor(np.arange(self.ni) / self.batch_size).astype(int)  # batch index
    nb = bi[-1] + 1  # number of batches

    s = np.array([x.pop("shape") for x in self.labels])  # hw
    ar = s[:, 0] / s[:, 1]  # aspect ratio
    irect = ar.argsort()
    self.im_files = [self.im_files[i] for i in irect]
    self.labels = [self.labels[i] for i in irect]
    ar = ar[irect]

    # Set training image shapes
    shapes = [[1, 1]] * nb
    for i in range(nb):
        ari = ar[bi == i]
        mini, maxi = ari.min(), ari.max()
        if maxi < 1:
            shapes[i] = [maxi, 1]
        elif mini > 1:
            shapes[i] = [1, 1 / mini]

    self.batch_shapes = np.ceil(np.array(shapes) * self.imgsz / self.stride + self.pad).astype(int) * self.stride
    self.batch = bi  # batch index of image

update_labels

update_labels(include_class: Optional[list])

Update labels to include only these classes (optional).

Source code in ultralytics/data/base.py
def update_labels(self, include_class: Optional[list]):
    """Update labels to include only these classes (optional)."""
    include_class_array = np.array(include_class).reshape(1, -1)
    for i in range(len(self.labels)):
        if include_class is not None:
            cls = self.labels[i]["cls"]
            bboxes = self.labels[i]["bboxes"]
            segments = self.labels[i]["segments"]
            keypoints = self.labels[i]["keypoints"]
            j = (cls == include_class_array).any(1)
            self.labels[i]["cls"] = cls[j]
            self.labels[i]["bboxes"] = bboxes[j]
            if segments:
                self.labels[i]["segments"] = [segments[si] for si, idx in enumerate(j) if idx]
            if keypoints is not None:
                self.labels[i]["keypoints"] = keypoints[j]
        if self.single_cls:
            self.labels[i]["cls"][:, 0] = 0

update_labels_info

update_labels_info(label)

Custom your label format here.

Source code in ultralytics/data/base.py
def update_labels_info(self, label):
    """Custom your label format here."""
    return label





Created 2023-11-12, Updated 2024-07-21
Authors: glenn-jocher (6), Burhan-Q (1)