You can get following advanteges:
(1) infinite training images generated on the fly
(2) customize image height and width
You can use it to:
(1) achieve a pre-trained model and then do transfer-learning
(2) check out the ability of some model on severe occlusion
Document:
(1) puppet.py main doc
(2) puppet_test.ipynb notebook for example
(3) eval.py evaluate mAP using OKS