Abstract
In this paper, we present a regression-based pose recognition method usingcascade Transformers. One way to categorize the existing approaches in thisdomain is to separate them into 1). heatmap-based and 2). regression-based. Ingeneral, heatmap-based methods achieve higher accuracy but are subject tovarious heuristic designs (not end-to-end mostly), whereas regression-basedapproaches attain relatively lower accuracy but they have less intermediatenon-differentiable steps. Here we utilize the encoder-decoder structure inTransformers to perform regression-based person and keypoint detection that isgeneral-purpose and requires less heuristic design compared with the existingapproaches. We demonstrate the keypoint hypothesis (query) refinement processacross different self-attention layers to reveal the recursive self-attentionmechanism in Transformers. In the experiments, we report competitive resultsfor pose recognition when compared with the competing regression-based methods.