MobilePose

MobilePose is a Tiny PyTorch implementation of single person 2D pose estimation framework. The aim is to provide the interface of the training/inference/evaluation, and the dataloader with various data augmentation options. And final trained model can satisfy basic requirements(speed+size+accuracy) for mobile device.

Some codes for networks and display are brought from:

1. pytorch-mobilenet-v2
2. Vanilla FCN, U-Net, SegNet, PSPNet, GCN, DUC
3. Shufflenet-v2-Pytorch
4. tf-pose-estimation
5. dsntnn

NEWS!

• Apr 2021: Siyuan Pan provides MNN version!
• Mar 2019: Support running on MacBook with decent FPS!
• Feb 2019: ALL the pretrained model files are avaliable!

Requirements

• Python 3.7
• PyTorch 1.0
• dsntnn 1.0

Evaluation Results

Features

• multi-thread dataloader with augmentations (dataloader.py)
• training and inference (training.py)
• performance evaluation (eval.py)
• multiple models support (network.py)
• ipython notebook visualization (demo.ipynb)
• Macbook camera realtime display script (run_webcam.py)

Usage

1. Installation:

pip install -r requirements.txt

2. Training:

python training.py --model shufflenetv2 --gpu 0 --inputsize 224 --lr 1e-3 --batchsize 128 --t7 ./models/shufflenetv2_224_adam_best.t7

3. Evaluation

ln -s cocoapi/PythonAPI/pycocotools
cd cocoapi/PythonAPI && make

python eval.py --t7 ./models/resnet18_224_adam_best.t7 --model resnet18 --gpu 0

4. Web Camera Demo (MacBook)

python run_webcam.py --model squeezenet --inp_dim 224 --camera 0

Contributors

MobilePose is developed and maintained by Yuliang Xiu, Zexin Chen and Yinghong Fang. Thanks for Siyuan Pan's implementation of mnn version.

License

MobilePose is freely available for free non-commercial use. For commercial queries, please contact Cewu Lu.