TFPose: Direct Human Pose Estimation with Transformers

TL;DR

TFPose introduces a transformer-based regression framework for human pose estimation, effectively addressing feature misalignment and leveraging keypoint relationships to outperform previous regression methods and match heatmap-based approaches.

Contribution

The paper presents a novel transformer-based regression approach for pose estimation that overcomes limitations of previous methods and achieves state-of-the-art results.

Findings

Significantly improves regression-based pose estimation accuracy.

Performs comparably with top heatmap-based methods.

Demonstrates effectiveness on MS-COCO and MPII datasets.

Abstract

We propose a human pose estimation framework that solves the task in the regression-based fashion. Unlike previous regression-based methods, which often fall behind those state-of-the-art methods, we formulate the pose estimation task into a sequence prediction problem that can effectively be solved by transformers. Our framework is simple and direct, bypassing the drawbacks of the heatmap-based pose estimation. Moreover, with the attention mechanism in transformers, our proposed framework is able to adaptively attend to the features most relevant to the target keypoints, which largely overcomes the feature misalignment issue of previous regression-based methods and considerably improves the performance. Importantly, our framework can inherently take advantages of the structured relationship between keypoints. Experiments on the MS-COCO and MPII datasets demonstrate that our method can significantly improve the state-of-the-art of regression-based pose estimation and perform comparably with the best heatmap-based pose estimation methods.