PE-former: Pose Estimation Transformer

Vision transformer architectures have been demonstrated to work very effectively for image classification tasks. Efforts solve more challenging vision tasks with transformers rely on convolutional backbones feature extraction. In this paper we investigate the use of a pure architecture (i.e., one no CNN backbone) problem 2D body pose estimation. We evaluate two ViT COCO dataset. demonstrate that using an encoder-decoder yields state art results estimation problem.