Gaussian Masked Autoencoders

Jathushan Rajasegaran, Xinlei Chen, Ruilong Li, Christoph Feichtenhofer, Shiry Ginosar, Jitendra Malik

Keywords: Representation learning, Gaussian Splatting

Abstract: This paper explores Masked Autoencoders (MAE) with Gaussian Splatting. While mainstream self-supervised learning frameworks such as MAE operate on low-level pixels, the image synthesis community has evolved to use latent, mid-level representations for better generative visual data modeling. Our approach, named GMAE, aims to reconcile these two and get the benefits of both worlds. Like MAE, it reconstructs the image end-to-end in the pixel space; however, it also introduces an intermediate, 3D Gaussian-based representation and renders images via splatting. We show that GMAE can enable various zero-shot learning capabilities (e.g figure-ground segmentation, image layering, edge detection, etc) while preserving the high self-supervised representation quality from MAE. Notably, we are the first to employ Gaussian primitives in an image representation learning framework beyond optimization-based single-scene reconstructions. We believe GMAE will inspire further research in this direction and contribute to developing next-generation techniques for modeling high-fidelity visual data.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

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: 12248