Publication
Particle-based 6D Object Pose Estimation from Point Clouds using Diffusion Models
Christian Möller; Niklas Funk; Jan Peters
In: Computing Research Repository eprint Journal (CoRR), Vol. abs/2412.00835, Pages 1-13, arXiv, 2024.
Abstract
Object pose estimation from a single view remains a chal-
lenging problem. In particular, partial observability, oc-
clusions, and object symmetries eventually result in pose
ambiguity. To account for this multimodality, this work pro-
poses training a diffusion-based generative model for 6D
object pose estimation. During inference, the trained gen-
erative model allows for sampling multiple particles, i.e.,
pose hypotheses. To distill this information into a single
pose estimate, we propose two novel and effective pose se-
lection strategies that do not require any additional training
or computationally intensive operations. Moreover, while
many existing methods for pose estimation primarily fo-
cus on the image domain and only incorporate depth in-
formation for final pose refinement, our model solely op-
erates on point cloud data. The model thereby leverages
recent advancements in point cloud processing and oper-
ates upon an SE(3)-equivariant latent space that forms the
basis for the particle selection strategies and allows for
improved inference times. Our thorough experimental re-
sults demonstrate the competitive performance of our ap-
proach on the Linemod dataset and showcase the effective-
ness of our design choices. Code is available at https:
//github.com/zitronian/6DPoseDiffusion.