How could I run a demo wild video to see the visualization result

[hujb48]

It‘s a great work! And can you upload the The corresponding pre-trained model to evaluate any wild video and to see the visualization result if possible! Best wishes!

[Garfield-kh]

Thanks for your interest! Here are the steps to predict 3D pose (16 joints) from image/video:

1. Apply OpenPose or AlphaPose to get the detect 2D.
2. Then convert the detect 2D to 16 joints definition with a similar procedure as prepare_data_3dhp.py .
3. Apply Savgol filter to smooth the 2D pose.
4. Preprocess the 2D pose with the same normalization as the 2D-3D pose model training.
5. Apply the 2D-3D pose model on 2D pose to get the 3D pose result.
6. Visualization the 3D pose result with code visualization.py

We are busy with other stuffs currently, we will update you once we have time to clean the code. Hope this helps. Thank you.

[hujb48]

@Garfield-kh Thanks a lot for your prompt reply, I am a new at this field, I would try it by following this guide, maybe it would have some problems with test and hope to communicate with your team at your convenience! Yours sincerely, Hu.

[Garfield-kh]

You are welcome~ :)

[hujb48]

Hi, there is another suggest that if you can give the pre-trained model trained with the trajectory model like https://github.com/facebookresearch/VideoPose3D, the model would be better for visualization intuitively. And Thanks again!

[Garfield-kh]

Thank you for the suggestion, will update you once we make it. ^-^

[hujb48]

:)

[JinShiyin]

Hi, the output is relative to the 0 joint, how can I get the 0 joint's 3d location?

[Garfield-kh]

Hi, you may check this 3DMPPE_ROOTNET_RELEASE or VideoPose3D root location prediction.

[JinShiyin]

Ok, thank you for your reply.