GitLab

* Creates a new batch_decode method in SSD Meta architecture that can handle
  dynamic batch size.
* use combined_shapes in _get_feature_maps_spatial_dims method to handle
  dynamic batch image_size.
* Add dynamic batch size tests to check preprocess, predict and postprocess
  methods in SSD Meta architecture.

* Adds a util function to compute a mix of dynamic and static shapes
  preferring static when available.
* Uses batch_multiclass_non_max_suppression function in postprocess_rpn
  instead of looping over static batch shape and performing
  multiclass_non_max_suppression.
* Adds a new helper function _unpad_proposals_and_sample_boxclassifier_batch
  to sample from a batch of tensors possibly containing paddings.
* Tests batch inference with various configurations of static shape via
  unittests.

A few change to prepare for batch inference:

* Modify the return type of batch non max suppression to be tuple of tensors
  so it can be reused for both stages of faster rcnn without any confusion
  in the semantics implied the the keys used to represent the tensors.
* Allow dynamic number of anchors (boxes) in addition to dynamic batch size.
* Remove a redundant dynamic batch size test.

This change is required to enable object detection export and inference
with dynamic batch size.

Bug:
We use two different Saver objects during training, one to initialize from a
checkpoint and another to save and restore from checkpoints during training.
Both these Savers were bound to same variable name causing init_fn function
to use the latter instead of the former when called.

Add back six dependency and change pbtxt files to reflect new way of using own data to train object recognition models

Correct documentation so all commands should be run from the root
of the git directory.

Also clarifies usage on label maps. In a nutshell, label map IDs
should start at index 1 (and not include 0).

* Change dictionnaries iteritems() to items()

Also use the url link to github because *.md files are not
rendered when using jupyter notebook

1 should not shuffle during evaluation; 2 num_readers should be one, or each sample would be evaluate for (num_reader) times and only (num_examples / num_reader) distinct samples are evaluted.