sellProject sellDFKI sellAugmented Vision

Category-Agnostic Pose Estimation

Topic

In this project, we will create a text-promptable pose estimation model that can be used to perform skeleton-agnostic human pose estimation. The goal is to dynamically modify the number of predicted keypoints at inference time in a zero-shot manner.

Tasks

1. Prepare a text-image dataset suitable for training a zero-shot CAPE model tailored for humans only​
2. Model implementation and training​
3. Detailed performance comparisons

Expected Skills

1. Strong programming skills + PyTorch (required) ​
2. Experience with human pose estimation (highly preferred) and MMPose (preferred)​

Related Literature

• [1] X-Pose
• [2] PoseAnything
• [3] CapeFormer