Keypoint augmentations not working for IAAAffine

🐛 Bug

As per the overview of transformations here: https://albumentations.ai/docs/getting_started/transforms_and_targets/ I try to run both pixel-wise and spatial augmentations on my dataset, however when using the IAAAffine transformation, it results in the following error: ValueError: Expected x for keypoint (-9.433, 93.50725, 0.0, 0) to be in the range [0.0, 256], got -9.433.

Would like to perform different translations/scaling/shearing of my dataset. Is this not possible with keypoints also? Or are there other native albumentations transforms that would work instead?

To Reproduce

Steps to reproduce the behavior:

augment = albu.Compose(
    [
        albu.HueSaturationValue(p=0.5),
        albu.RGBShift(p=0.5),
        albu.RandomBrightnessContrast(p=0.5),
        albu.MotionBlur(p=0.5),
        albu.Cutout(num_holes=6, max_h_size=ranh, max_w_size=ranw, fill_value=ranc, always_apply=False, p=0.5),
        albu.IAAAffine(scale=1, translate_percent=0.0, translate_px=None, rotate=0.0, shear=[25,25], order=1, cval=0, mode='edge', always_apply=False, p=0.5),
        albu.GaussNoise(p=0.5),
        albu.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225), max_pixel_value=255.0, always_apply=False, p=1.0),
    ],
    keypoint_params=albu.KeypointParams(format="xy")
)

augmented = augment(image=image, mask=mask, keypoints=keypoints)

Environment

• Albumentations version (e.g., 0.1.8): 0.5.2
• Python version (e.g., 3.7): 3.7.9
• OS (e.g., Linux): Windows
• How you installed albumentations (conda, pip, source): conda

[Dipet]

The problem is not in imagaug augmentations. You have keypoints outside of image. If you want to use and save them set remove_invisible=False

keypoint_params=albu.KeypointParams(format="xy", remove_invisible=False)

[Dipet]

Sorry for the delay. How did you get this white background? I can not reproduce the problem. By default set border_mode=cv2.BORDER_REFLECT_101, which means that all pixels outside the image will be reflected starting from the image edge.