/centerpose

Push the Extreme of the pose estimation

Primary LanguagePythonMIT LicenseMIT

The repo is based on CenterNet, which aimed for push the boundary of human pose estimation

multi person pose estimation using center point detection:

Main results

Keypoint detection on COCO validation 2017

Backbone AP FPS TensorRT Speed Download
DLA-34 62.7 23 - model
Resnet-50 54.5 28 33 model
MobilenetV3 46.0 30 - model
ShuffleNetV2 40.7 25 - model
High Resolution 49.5 16 - model
HardNet 45.6 30 - model
Darknet53 33.2 30 - model

Installation

git submodule init&git submodule update Please refer to INSTALL.md for installation instructions.

Use CenterNet

We support demo for image/ image folder, video, and webcam.

First, download the model DLA-34 from the Model zoo and put them in anywhere.

Run:

cd tools; python demo.py --cfg ../experiments/dla_34_512x512.yaml --TESTMODEL /your/model/path/dla34_best.pth --DEMOFILE ../images/33823288584_1d21cf0a26_k.jpg --DEBUG 1

The result for the example images should look like:

Training

After installation, follow the instructions in DATA.md to setup the datasets.

We provide config files for all the experiments in the experiments folder.

cd ./tools python -m torch.distributed.launch --nproc_per_node 4 train.py --cfg ../experiments/*yalm

License

MIT License (refer to the LICENSE file for details).

Citation

If you find this project useful for your research, please use the following BibTeX entry.

@inproceedings{zhou2019objects,
  title={Objects as Points},
  author={Zhou, Xingyi and Wang, Dequan and Kr{\"a}henb{\"u}hl, Philipp},
  booktitle={arXiv preprint arXiv:1904.07850},
  year={2019}
}