UKPGAN: A General Self-Supervised Keypoint Detector

TL;DR

UKPGAN is a novel self-supervised 3D keypoint detection method that identifies important points for shape reconstruction, demonstrating strong alignment with human labels and robustness across transformations.

Contribution

It introduces a general self-supervised framework with GAN-based sparsity control and salient information distillation for 3D keypoint detection, applicable to various objects and scenarios.

Findings

Keypoints align well with human annotations.

Effective on non-rigid deformations like SMPL models.

Improves geometric registration with off-the-shelf descriptors.

Abstract

Keypoint detection is an essential component for the object registration and alignment. In this work, we reckon keypoint detection as information compression, and force the model to distill out irrelevant points of an object. Based on this, we propose UKPGAN, a general self-supervised 3D keypoint detector where keypoints are detected so that they could reconstruct the original object shape. Two modules: GAN-based keypoint sparsity control and salient information distillation modules are proposed to locate those important keypoints. Extensive experiments show that our keypoints align well with human annotated keypoint labels, and can be applied to SMPL human bodies under various non-rigid deformations. Furthermore, our keypoint detector trained on clean object collections generalizes well to real-world scenarios, thus further improves geometric registration when combined with off-the-shelf point descriptors. Repeatability experiments show that our model is stable under both rigid and non-rigid transformations, with local reference frame estimation. Our code is available on https://github.com/qq456cvb/UKPGAN.