Introductioon
Unofficial implementation of MoCap-guided Data Augmentation for 3D Pose Estimation in the Wild (NeurIPS 2016). Just for fun! This repo does not provide the way to reproduce the paper. This repo utilizes the method to convert Human3.6M images to MSCOCO style.
How to run
Prepare Human3.6M and MSCOCO data. You can refer to here.
Under ${ROOT}/tool/, run python mosaic.py --subject 1 --part 1. subject is Human3.6M subject. part is the partition number.
Change the output directory in line 50 of mosaic.py before running.
You may improve the quality by commenting line 147,148 of MSCOCO.py.
Results
Good cases
Awkward cases
Bad cases
