vitpose_deploy
Copyright 2022 The University of Sydney.
Copyright 2024 The HuggingFace Inc. team.

This product includes software developed at the University of Sydney, published
as ViTPose (https://github.com/ViTAE-Transformer/ViTPose) and described in
"ViTPose: Simple Vision Transformer Baselines for Human Pose Estimation" (Xu,
Zhang, Zhang, Tao; NeurIPS 2022, arXiv:2204.12484) and its extension "ViTPose++:
Vision Transformer for Generic Body Pose Estimation" (TPAMI 2023,
arXiv:2212.04246).

The authors' own release is built on mmpose (OpenMMLab, Apache-2.0). The modules
here are extracted from the PyTorch implementation that HuggingFace publishes as
transformers/models/vitpose and transformers/models/vitpose_backbone, which is
licensed under the Apache License, Version 2.0; see the LICENSE file in this
directory, which is the Apache-2.0 text those projects publish. HuggingFace
prepends its own copyright line to that file; the line is reproduced above rather
than inside the licence text, which is where the other extractions here keep it.

The published checkpoints are Apache-2.0 as well -- all seven of them, code and
model weights under the same terms. Nothing in this family carries a separate
weights licence.

The joint names in the published labels.json are COCO's own, from the
person_keypoints annotations (Creative Commons Attribution 4.0). Seventeen names
for seventeen channels is not a creative work, and they are reproduced here
because they are what the checkpoint's head was trained against.

NOT INCLUDED
------------
Training, evaluation and the flip-test augmentation are not part of this package.
Neither is the simple decoder, nor the single-expert MLP that the original
ViTPose used -- no published variant here takes either branch. See PROVENANCE.md
for the full list and for why.
