rishitdagli/objaverse-renders

This dataset was used in the paper Squeeze3D: Extreme Neural Compression with Latent Space Bridging (TMLR 2026). It contains 10,000 renders of Objaverse objects, split across part1/ and part2/, that were used to create the InstantMesh and OpenLRM mesh datasets. The renders were created using G-buffer Objaverse.

Paper: https://openreview.net/forum?id=XXYGHfqzvA

Project page: https://squeeze3d.github.io/

Code: https://github.com/Rishit-dagli/Squeeze3D

Citation

@article{dagli2026squeeze3d,
title={Squeeze3D: Extreme Neural Compression with Latent Space Bridging},
author={Rishit Dagli and Yushi Guan and Sankeerth Durvasula and Mohammadreza Mofayezi and Nandita Vijaykumar},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2026},
url={https://openreview.net/forum?id=XXYGHfqzvA}
}