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.

Replace deprecated get_or_create_global_step

update broken ssd models in the zoo along with notebook.

minor updates in documentation.

Add dataset tools, pretrained models and documentatinos for open imag…

add inference tools for Open Image dataset.

update offline metrics for Open Images dataset.

move dataset tools into a subfolder.

* Add simplifications requested by get_started writers.

* More cleanup

* Clearer names and comments

* Use the keys from the train dict

* Use regular input_fn pattern

* Add comments to input functions

Update blog_estimators_dataset.py requires r1.4 for tf.data

Added assert for TensorFlow r1.4 since tf.data is required.

Add autograd v1.1.13 dependency to manual install

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

* Release DRAGNN bulk networks

Fix SyntaxNet/DRAGNN Dockerfile build

Attention OCR: Add missing is_training to batch_norm and dropout

#attention_ocr: fix deprecation warnings and update usage examples

Consistent, fixed-width printing from adversarial crypto example.

correct faster rcnn nas link.

fix preprocessor builder to pass correct kwargs to random_crop_pad op.

update ssd anchor generator and model zoo