Tensorflow Object Detection (TFOD), data augmentation error: image_size must contain 3 elements

Question

I use a TFOD object detection framework, and visualize all data augmentation functions. There are several functions, that throw the exception:

InvalidArgumentError: image_size must contain 3 elements[1] [Op:SampleDistortedBoundingBoxV2]

These functions are related to SSD augmentations:

• SSD random crop
• SSD random crop pad
• SSD random crop fixed aspect ratio

Previously, the guys (https://github.com/tensorflow/models/issues/3349) advised to check if the image is RGB. I copied the code and made sure - image is RGB. Shape of the image - tf.Tensor([320, 320, 3]). So, the image_size has exactly 3 elements.

Before, i normalize the image, because it is requested that pixels are from 0 to 1 range.

Here is the short snippet of a big system.

def modify(self, image_):
    
    # before ssd, pixels must be [0, 1]
    normalized_img = preprocessor.normalize_image(
        image_,
        original_minval=0,
        original_maxval=255,
        target_minval = 0,
        target_maxval = 1
    )

    
    assert np.max(normalized_img.numpy()) <= 1.0
    assert np.min(normalized_img.numpy()) >= 0.0

    modified_img = self.augmentation_function(
        normalized_img, **self._augmentation_parameters
    )
    return modified_img

Why do these functions continue requesting 3 elements in the image size? Thank you so much!