ControLRM: Fast and Controllable 3D Generation via Large Reconstruction Model

TL;DR

ControLRM is an end-to-end, fast, and controllable 3D generation model that leverages large reconstruction models and joint training to improve alignment and efficiency over traditional methods.

Contribution

We introduce ControLRM, a novel end-to-end model that enables rapid and controllable 3D generation with a joint training framework and implicit 2D-3D alignment.

Findings

Demonstrates strong generalization across multiple datasets.

Achieves superior controllability and quality in 3D generation.

Reduces training time compared to score-distillation sampling methods.

Abstract

Despite recent advancements in 3D generation methods, achieving controllability still remains a challenging issue. Current approaches utilizing score-distillation sampling are hindered by laborious procedures that consume a significant amount of time. Furthermore, the process of first generating 2D representations and then mapping them to 3D lacks internal alignment between the two forms of representation. To address these challenges, we introduce ControLRM, an end-to-end feed-forward model designed for rapid and controllable 3D generation using a large reconstruction model (LRM). ControLRM comprises a 2D condition generator, a condition encoding transformer, and a triplane decoder transformer. Instead of training our model from scratch, we advocate for a joint training framework. In the condition training branch, we lock the triplane decoder and reuses the deep and robust encoding layers pretrained with millions of 3D data in LRM. In the image training branch, we unlock the triplane decoder to establish an implicit alignment between the 2D and 3D representations. To ensure unbiased evaluation, we curate evaluation samples from three distinct datasets (G-OBJ, GSO, ABO) rather than relying on cherry-picking manual generation. The comprehensive experiments conducted on quantitative and qualitative comparisons of 3D controllability and generation quality demonstrate the strong generalization capacity of our proposed approach.