__array_interface__ support?

[radarhere]

Click to expand!

Issue Type

Feature Request

Source

source

Tensorflow Version

2.6.0

Custom Code

No

OS Platform and Distribution

No response

Mobile device

No response

Python version

3.8

Bazel version

No response

GCC/Compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current Behaviour?

Hi. Over at https://github.com/python-pillow/Pillow, our Image class has the __array_interface__ attribute, to support converting Pillow images to NumPy. See https://numpy.org/doc/stable/reference/arrays.interface.html#object.__array_interface__

A recent discussion has revealed that TensorFlow's reshape method (and I have to imagine other methods as well) accepts an object that provides an __array__ method, but not an object with an __array_interface__ attribute.

My question - is there any interest from TensorFlow in supporting objects with __array_interface__?

Standalone code to reproduce the issue

Here is code that fails with the latest version of Pillow.

from PIL import Image
import tensorflow as tf

im = Image.new("L", (1, 1))
tf.reshape(im, (1, 1))

However, if my assertion that TensorFlow doesn't accept __array_interface__ is at all in doubt, let me know and I'll put together a better example.

### Relevant log output

_No response_</details>

[chunduriv]

The issue is replicated in TF 2.9

from PIL import Image
import tensorflow as tf

im = Image.new("L", (1, 1))
tf.reshape(im, (1, 1))

Output:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
[<ipython-input-1-807613033b71>](https://localhost:8080/#) in <module>()
      3 
      4 im = Image.new("L", (1, 1))
----> 5 tf.reshape(im, (1, 1))

1 frames
[/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py](https://localhost:8080/#) in convert_to_eager_tensor(value, ctx, dtype)
    100       dtype = dtypes.as_dtype(dtype).as_datatype_enum
    101   ctx.ensure_initialized()
--> 102   return ops.EagerTensor(value, ctx.device_name, dtype)
    103 
    104 

ValueError: Attempt to convert a value (<PIL.Image.Image image mode=L size=1x1 at 0x7F4CA06B7090>) with an unsupported type (<class 'PIL.Image.Image'>) to a Tensor.

[radarhere]

Thanks very much @cantonios!

[cantonios]

@radarhere no problem. At Google we oddly have a patched version of Pillow that replaces __array_interface__ back with __array__, so we weren't reproducing this issue internally. Maybe this was why.