Fix crash of tf.image.extract_glimpse with negative input

[yongtang]

This PR tries to fix the issue raised in #51618 where
tf.image.extract_glimpse will crash in case of negative input.

This PR adds additional checking to allow graceful error message.

This PR fixes #51618.

Signed-off-by: Yong Tang yong.tang.github@outlook.com

[jpienaar]

Nice, thanks!

[gbaned]

Here are the internal errors, @yongtang can you please verify ? Thanks!

Traceback (most recent call last):
File "/tensorflow/python/framework/ops.py", line 1926, in _create_c_op
c_op = pywrap_tf_session.TF_FinishOperation(op_desc)
google3.third_party.tensorflow.python.framework.errors_impl.InvalidArgumentError: Shape must be rank 2 but is rank 1 for '{{node ExtractGlimpseV2}} = ExtractGlimpseV2[centered=false, noise="uniform", normalized=false, uniform_noise=false](ExtractGlimpseV2/input, ExtractGlimpseV2/size, ExtractGlimpseV2/offsets)' with input shapes: [1,3,3,1], [2], [2].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/tensorflow/python/kernel_tests/attention_ops_test.py", line 314, in testGlimpseNegativeInput
normalized=False)
File "/tensorflow/python/util/traceback_utils.py", line 141, in error_handler
return fn(*args, **kwargs)
File "/tensorflow/python/util/dispatch.py", line 1048, in op_dispatch_handler
return dispatch_target(*args, **kwargs)
File "/tensorflow/python/ops/image_ops_impl.py", line 4945, in extract_glimpse_v2
name=name)
File "/tensorflow/python/ops/gen_image_ops.py", line 2015, in extract_glimpse_v2
name=name)
File "/tensorflow/python/framework/op_def_library.py", line 751, in _apply_op_helper
attrs=attr_protos, op_def=op_def)
File "/tensorflow/python/framework/ops.py", line 3695, in _create_op_internal
op_def=op_def)
File "/tensorflow/python/framework/ops.py", line 2088, in init
control_input_ops, op_def)
File "/tensorflow/python/util/traceback_utils.py", line 141, in error_handler
return fn(*args, **kwargs)
File "/tensorflow/python/framework/ops.py", line 1929, in _create_c_op
raise ValueError(e.message)
ValueError: Shape must be rank 2 but is rank 1 for '{{node ExtractGlimpseV2}} = ExtractGlimpseV2[centered=false, noise="uniform", normalized=false, uniform_noise=false](ExtractGlimpseV2/input, ExtractGlimpseV2/size, ExtractGlimpseV2/offsets)' with input shapes: [1,3,3,1], [2], [2].

[yongtang]

Thanks @mihaimaruseac @gbaned, I have updated the PR with additional exceptions captured. Please give it a try.