GitLab

Before the fix the demo_inference.py used batch_norm and it did the
normalization of input image implicitly. If at inference time the
batch_norm was disabled the inference produced incorrect results.
This fix does the proper input image normalization and disables the batch_norm
at inference time.

tf.contrib.framework.get_or_create_global_step -> tf.train.get_or_create_global_step

* Release DRAGNN bulk networks

* add configs corresponding to new models in the zoo.
* add warinings aboud version in notebook.

add interpolated scale at a give aspect ratio. This change breaks
the existing checkpoints because the anchor orders are different.

This change should be merged along newly generated frozen ssd models.

- add patch package to make bazel work with @opt
- CI configured is trying to get git version, which lead to fail since the parents folders are not copied
- bazel version was too old for the new tensorflow version
- bazel not finding PYTHON_BIN_PATH, add ENV command as a workaround

Fix nasnet image classification and object detection by moving the
option to turn ON or OFF batch norm training into it's own arg_scope
used only by detection

PiperOrigin-RevId: 173917628

PiperOrigin-RevId: 173823563

Add __init__.py file in nasnet directory.

Remove google3 dependency on opensourced NASNet.