FlexPose: Pose Distribution Adaptation with Limited Guidance

TL;DR

FlexPose is a method for adapting pre-trained human pose generators to new pose distributions using limited guidance, enabling efficient transfer learning across datasets with different pose priors.

Contribution

The paper introduces FlexPose, a novel approach that fine-tunes a pre-trained pose generator with minimal annotation guidance by focusing on pose transformation layers.

Findings

Achieves state-of-the-art transfer performance with limited guidance.

Effectively adapts to diverse pose distributions across datasets.

Reduces annotation effort in pose dataset creation.

Abstract

Numerous well-annotated human key-point datasets are publicly available to date. However, annotating human poses for newly collected images is still a costly and time-consuming progress. Pose distributions from different datasets share similar pose hinge-structure priors with different geometric transformations, such as pivot orientation, joint rotation, and bone length ratio. The difference between Pose distributions is essentially the difference between the transformation distributions. Inspired by this fact, we propose a method to calibrate a pre-trained pose generator in which the pose prior has already been learned to an adapted one following a new pose distribution. We treat the representation of human pose joint coordinates as skeleton image and transfer a pre-trained pose annotation generator with only a few annotation guidance. By fine-tuning a limited number of linear layers that closely related to the pose transformation, the adapted generator is able to produce any number of pose annotations that are similar to the target poses. We evaluate our proposed method, FlexPose, on several cross-dataset settings both qualitatively and quantitatively, which demonstrates that our approach achieves state-of-the-art performance compared to the existing generative-model-based transfer learning methods when given limited annotation guidance.