make sure batch_norm is False when inferencing a image

[beacandler]

problem

when using
code: tensorflow/models/reaserch/atttention_ocr:
paper: Attention-based Extraction of Structured Information from Street View Imagery
I encountered a issue.
Recognizition result for one image is depended on images within the same batch, sometimes when batch is larger than the number of acutal images ,the result is absurd.

reason

in batch_norm.
When inferencing a image in a batch, we expect attention_ocr use global mean and std to act on the image, but When Is_training = True(default), attention_ocr will get mean and std by computing in a batch, so the error occurs.

solution

pass is_training=False to batch_norm when inferencing.
so in attention_ocr/python/model.py 209:63
with slim.arg_scope([slim.batch_norm], is_training=is_training)

[alexgorban]

It looks like you are referring to an old version of the code. The issue with batch_norm was addressed in #3111 back in Nov 2017 (code). Please sync and let me know if you still experience the same problem.

[Harshini-Gadige]

Automatically closing due to lack of recent activity. Please update the issue when new information becomes available, and we will reopen the issue. Thanks!