You may check out the following tags for previous versions of this software.
Massively overhauled for Tensorflow 1.10.
This version uses the TensorFlow Dataset for fast I/O. Training, testing, validation, and prediction use a custom Estimator.
The input pipeline was refactored to account for a dynamically
generated data source
(MapTextSynthesizer),
with supporting command-line flags were added to train.py
.
Updated for Tensorflow 1.8.
This version uses the original TensorFlow Reader and QueueRunner mechanisms for fast, parallel I/O. For training it uses a straightforward MonitoredTrainingSession. Testing and evaluation manually manage sessions and checkpoints.
Updated for Tensorflow 1.1/1.2.
This version uses the original TensorFlow Reader and QueueRunner mechanisms for fast, parallel I/O. For training it uses a Supervisor with a managed session. Testing and evaluation manually manage sessions and checkpoints.
Initial version for Tensorflow 1.1. Lacked validate.py
and a few
other bug fixes added to tag tf-1.2
.