An unofficial implementation of ViTPose [Y. Xu et al., 2022]

python inference.py --image-path './examples/img1.jpg'

python train.py --config-path config.yaml --model-name 'b'

• model_name must be in (b, l, h)

1. Download the trained model (.pth)
   • ViTPose-B-Multi-COCO.pth
   • ViTPose-L-Multi-COCO.pth
   • ViTPose-H-Multi-COCO.pth
2. Set the config. according to the trained model
   • ViTPose-B-COCO-256x192
   • ViTPose-L-COCO-256x192
   • ViTPose-H-COCO-256x192

All codes were written with reference to the official ViTPose repo.