Skip to content

deeplabcut.pose_estimation_tensorflow.datasets.utils

Functions:

Name Description
crop_image

Randomly cropping image around xlabel,ylabel taking into account size of image.

crop_image 

crop_image(joints, im, Xlabel, Ylabel, cfg)

Randomly cropping image around xlabel,ylabel taking into account size of image.

Introduced in DLC 2.0 (Nature Protocols paper)

Source code in deeplabcut/pose_estimation_tensorflow/datasets/utils.py 
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
def crop_image(joints, im, Xlabel, Ylabel, cfg):
    """Randomly cropping image around xlabel,ylabel taking into account size of image.

    Introduced in DLC 2.0 (Nature Protocols paper)
    """
    widthforward = int(cfg["minsize"] + np.random.randint(cfg["rightwidth"]))
    widthback = int(cfg["minsize"] + np.random.randint(cfg["leftwidth"]))
    hup = int(cfg["minsize"] + np.random.randint(cfg["topheight"]))
    hdown = int(cfg["minsize"] + np.random.randint(cfg["bottomheight"]))
    Xstart = max(0, int(Xlabel - widthback))
    Xstop = min(np.shape(im)[1] - 1, int(Xlabel + widthforward))
    Ystart = max(0, int(Ylabel - hdown))
    Ystop = min(np.shape(im)[0] - 1, int(Ylabel + hup))
    joints[0, :, 1] -= Xstart
    joints[0, :, 2] -= Ystart

    inbounds = np.where(
        (joints[0, :, 1] > 0)
        * (joints[0, :, 1] < np.shape(im)[1])
        * (joints[0, :, 2] > 0)
        * (joints[0, :, 2] < np.shape(im)[0])
    )[0]
    return joints[:, inbounds, :], im[Ystart : Ystop + 1, Xstart : Xstop + 1, :]