Can this be trained with just the MS COCO dataset? - Githubissues

Daniil-Osokin / lightweight-human-pose-estimation-3d-demo.pytorch

Real-time 3D multi-person pose estimation demo in PyTorch. OpenVINO backend can be used for fast inference on CPU.

Apache License 2.0

656 stars 138 forks source link

Can this be trained with just the MS COCO dataset? #29

Closed delebash closed 4 years ago

delebash commented 4 years ago

The MS COCO is open source where the CMU Panoptic is for research or non-profit use only

Daniil-Osokin commented 4 years ago

Hi! To train 3D pose estimation a dataset with 3D keypoints annotations is needed. MS COCO has only 2D annotations, so 3D dataset is needed.

delebash commented 4 years ago

Thanks

delebash commented 4 years ago

Would this open source dataset work? https://labicvl.github.io/APE.html

Daniil-Osokin commented 4 years ago

Think so, it has 3D joints annotations.

delebash commented 4 years ago

Great how do I train this?

Daniil-Osokin commented 4 years ago

Check #30.

Daniil-Osokin commented 4 years ago

Hope, it is clear now.

delebash commented 4 years ago

Thanks