[pls merge] openpose example

[zsdonghao]

This is the TensorFlow implementation of "Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields"

There are some extra works need to be done before we can merge this PR.

• APIs for evaluation
• APIs for deployment (TensorRT)
• Provides pre-trained model here: https://github.com/tensorlayer/pretrained-models/tree/master/models
• Provides averaged precision (AP) results of the pre-trained model in readme.
• Moves (numpy-based, non-TF) data augmentation APIs into tl.prepro.
• Moves COCO dataset download and preprocess code into tl.files that allow users to automatically download and prepare the dataset.
• Readme:
  • introduction of this project
  • how to use your pre-trained model
  • how to download dataset
  • how to train and evaluate model
  • how to train model on customised dataset
  • how to customise data augmentation

To speed up the training:

• use tl.distributed.Trainer

To speed up the inferencing:

• For pose-processing, uses the C implementation of OpenPose
• Smaller model
• TensorRT float16 (half-float) inferencing

For TensorLayer Project Quality:

• Update Changelog.md
• Add unittests