Add MPI-INF-3DHP to datasets/ (3D human body tracking)

tensorflow / graphics

TensorFlow Graphics: Differentiable Graphics Layers for TensorFlow

Apache License 2.0

2.75k stars 364 forks source link

Add MPI-INF-3DHP to datasets/ (3D human body tracking) #313

Open taiya opened 4 years ago

taiya commented 4 years ago

This is a human tracking dataset collected by the MPI multi-view setup. It contains images + 3D marker locations (AFAIK)

PDF: https://arxiv.org/abs/1611.09813

@misc{mehta2016monocular,
    title={Monocular 3D Human Pose Estimation In The Wild Using Improved CNN Supervision},
    author={Dushyant Mehta and Helge Rhodin and Dan Casas and Pascal Fua and Oleksandr Sotnychenko and Weipeng Xu and Christian Theobalt},
    year={2016}}

taiya commented 4 years ago

Potential baselines associated with this dataset (might branch out to separate task): "HMR" → Kanazawa et al. "End-to-end recovery of human shape and pose"

(removed some comments below, as the discussion diverged)

ghosalsattam commented 4 years ago

@taiya Looks interesting. I am planning to work on it, but am unable to decide how to start. Can you help me? Shall we implement it in steps? Shall we import PosNet from tensorflow-models/posnet? Where shall we put the project(in tensorflow_graphics/projects or somewhere else)? Shall we add different functionalities as different pull request or just one PR with very large diff size? Sorry, if it is a common issue, but I am a newlearner(earlier I addressed a small issue) and found the project interesting. Thank you.