search

una-dinosauria / 3d-pose-baseline

A simple baseline for 3d human pose estimation in tensorflow. Presented at ICCV 17.
MIT License
1.41k stars 354 forks source link

How cameras.h5 file is generated? #146

Closed ML-DL-AI-SS closed 4 years ago

ML-DL-AI-SS commented 4 years ago

How the cameras.h5 file is being generated ? is it the data from D3_positions_mono data from human 3.6m?

Thanks in advance!!

una-dinosauria commented 4 years ago

We saved the camera parameters used in the H3.6M code to generated 2d projections from 3d inputs.

una-dinosauria commented 4 years ago

Closing for lack of activity.

Adilrn commented 3 years ago

@una-dinosauria how can I get the cameras.h5 since the link to get the data was removed? I was able to get the h3.6m D3_position but without the cameras.5, I am not able to run your repository

una-dinosauria commented 3 years ago

@AdRever cameras.h5 was generated by parsing the metadata.xml file from the Human 3.6M software v1.2 (available on the H3.6M website); there is a <projectionParams> section that contains the camera parameters. Unfortunately, I cannot distribute that file anymore, since that violates the Human3.6M licence agreement.

I've started working on a branch that reads the data from the format available directly on the website: https://github.com/una-dinosauria/3d-pose-baseline/tree/cdf, and I'm only missing generating the cameras.h5 file. Should be done soon.

Adilrn commented 3 years ago

Thank you. I will try to generate that while waiting for your new release.

una-dinosauria commented 3 years ago

I merged the changes last night. Please feel free to have a look.