Improving Viewpoint Consistency in 3D Generation via Structure Feature and CLIP Guidance

TL;DR

This paper introduces a tuning-free method called ACG that improves viewpoint consistency in text-to-3D generation by controlling attention, filtering viewpoints with CLIP, and refining through staged prompts, effectively reducing the Janus Problem.

Contribution

The paper proposes a novel, tuning-free ACG mechanism that enhances viewpoint accuracy in 3D generation without additional training, addressing the Janus Problem.

Findings

Significantly reduces the Janus Problem in 3D generation.

Maintains high generation speed while improving viewpoint consistency.

Serves as an efficient plug-and-play component for existing frameworks.

Abstract

Despite recent advances in text-to-3D generation techniques, current methods often suffer from geometric inconsistencies, commonly referred to as the Janus Problem. This paper identifies the root cause of the Janus Problem: viewpoint generation bias in diffusion models, which creates a significant gap between the actual generated viewpoint and the expected one required for optimizing the 3D model. To address this issue, we propose a tuning-free approach called the Attention and CLIP Guidance (ACG) mechanism. ACG enhances desired viewpoints by adaptively controlling cross-attention maps, employs CLIP-based view-text similarities to filter out erroneous viewpoints, and uses a coarse-to-fine optimization strategy with staged prompts to progressively refine 3D generation. Extensive experiments demonstrate that our method significantly reduces the Janus Problem without compromising generation speed, establishing ACG as an efficient, plug-and-play component for existing text-to-3D frameworks.