ZheC / Realtime_Multi-Person_Pose_Estimation

Code repo for realtime multi-person pose estimation in CVPR'17 (Oral)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Failed cases

albertchristianto opened this issue · comments

Hi,
Have you tried your COCO trained model to inference image like below ?
000533914
007787006
I have no idea why your model didn't output any human pose in that image. Or I need to fine tune the model.
Thank you for your attention.

People inside water and upside-down people are data that has not been given into the OpenPose training code. Your only change is to fine-tune it with those kind of samples.

For the upside-down people, you can also rotate the image 180deg when giving it to OpenPose (there is a flag to rotate the image)