search

vchoutas / smplify-x

Expressive Body Capture: 3D Hands, Face, and Body from a Single Image
https://smpl-x.is.tue.mpg.de/
Other
1.77k stars 340 forks source link

Suggested way to generate keypoints? #1

Closed lolz0r closed 5 years ago

lolz0r commented 5 years ago

Hi, Very cool work! If a person wanted to try this on their own imageset how should we generate the required keypoints?

Best wishes, Michael

geopavlakos commented 5 years ago

In our experiments, we used the OpenPose library (https://github.com/CMU-Perceptual-Computing-Lab/openpose) to generate keypoints. The code is also compatible with the format that OpenPose uses to store detections, so it should be straightforward to first run OpenPose and then use its output to apply our fitting code.

wangsen1312 commented 5 years ago

what kind of openpose versio should we use? BODY_25 or COCO

geopavlakos commented 5 years ago

The default version is BODY_25. For best results when fitting the SMPL-X model, we also suggest extracting hand and face keypoints, using the appropriate flags.