DC3DO: Diffusion Classifier for 3D Objects

Nursena Koprucu, Meher Shashwat Nigam, Shicheng Luke Xu, Biruk Abere Ambaw, Andrew Rodriguez, Gabriele Dominici, Sharvaree Vadgama, Berfin Inal, Alberto Tono

Keywords: diffusion, classifier, 3D, deep generative models, classification

TL;DR: 3D Classification with Diffusion Models Compared With Multi-View 2D Diffusion Classifier

Abstract: Recent advancements in deep generative models, particularly diffusion models, have shown remarkable capabilities in generating high-fidelity 3D objects. In this work, we explore the application of diffusion models for 3D object classification by integrating the LION model with diffusion-based classifiers. Due to the availability of pretrained model weights, our study focuses on two categories from the ShapeNet dataset: chairs and cars. We propose DC3DO, a method that leverages the generative strengths of diffusion models for domain generalization in 3D classification tasks. Our approach demonstrates improved performance over a multi-view baseline, highlighting the potential of diffusion models in handling 3D data. We also examine the model's ability to generalize to data from different distributions, evaluating its performance on the IFCNet and ModelNet datasets. This study underscores the potential of using diffusion models for 3D object classification and sets the stage for future research involving more categories as resources become available.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7869