tfa.image.transform interpolates pixels that are outside image boundary (instead of using `fill_value`)

[eli-osherovich]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Debian testing
• TensorFlow version and how it was installed (source or binary): 2.4.0 (binary)
• TensorFlow-Addons version and how it was installed (source or binary): 0.12.0 (binary)
• Python version: 3.8
• Is GPU used? (yes/no): no

Describe the bug

tfa.image.transform interpolates pixels that are outside image boundary (instead of using fill_value)
In the example below, the corner pixels are mapped from coordinates that lie outside the image. Hence, they must be set to fill_value (like, for example, scipy.ndimage.affine_transform and skimage.transform.AffineTransform). However, they are interpolated.

Code to reproduce the issue

import numpy as np
import tensorflow as tf
import tensorflow_addons as tfa

a = np.ones((3,3))
tfa.image.rotate(a, np.deg2rad(45), interpolation="bilinear")

<tf.Tensor: shape=(3, 3), dtype=float64, numpy=
array([[0.58578646, 1.        , 0.58578634],
       [1.        , 1.        , 1.        ],
       [0.58578646, 1.        , 0.58578634]])>

Other info / logs

Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

[WindQAQ]

We invoke tf.raw_ops.ImageProjectiveTransformV3 so it's better to post the feature request in core TensorFlow.

[eli-osherovich]

Opened the issue in tensorflow: tensorflow/tensorflow#46637