VOC2012¶

class paddle.vision.datasets. VOC2012 ( data_file=None, mode='train', transform=None, download=True, backend=None ) [source]

Implementation of VOC2012 dataset.

Parameters

• data_file (str, optional) – Path to data file, can be set None if download is True. Default: None, default data path: ~/.cache/paddle/dataset/voc2012.

• mode (str, optional) – Either train or test mode. Default ‘train’.

• transform (Callable, optional) – Transform to perform on image, None for no transform. Default: None.

• download (bool, optional) – Download dataset automatically if data_file is None. Default: True.

• backend (str, optional) – Specifies which type of image to be returned: PIL.Image or numpy.ndarray. Should be one of {‘pil’, ‘cv2’}. If this option is not set, will get backend from paddle.vision.get_image_backend, default backend is ‘pil’. Default: None.

Returns

Dataset. An instance of VOC2012 dataset.

Examples

import itertools
import paddle.vision.transforms as T
from paddle.vision.datasets import VOC2012


voc2012 = VOC2012()
print(len(voc2012))
# 2913

for i in range(5):  # only show first 5 images
    img, label = voc2012[i]
    # do something with img and label
    print(type(img), img.size)
    # <class 'PIL.JpegImagePlugin.JpegImageFile'> (500, 281)
    print(type(label), label.size)
    # <class 'PIL.PngImagePlugin.PngImageFile'> (500, 281)


transform = T.Compose(
    [
        T.ToTensor(),
        T.Normalize(
            mean=[0.5, 0.5, 0.5],
            std=[0.5, 0.5, 0.5],
            to_rgb=True,
        ),
    ]
)

voc2012_test = VOC2012(
    mode="test",
    transform=transform,  # apply transform to every image
    backend="cv2",  # use OpenCV as image transform backend
)
print(len(voc2012_test))
# 1464

for img, label in itertools.islice(iter(voc2012_test), 5):  # only show first 5 images
    # do something with img and label
    print(type(img), img.shape)
    # <class 'paddle.Tensor'> [3, 281, 500]
    print(type(label), label.shape)
    # <class 'numpy.ndarray'> (281, 500)