X-Oscar: A Progressive Framework for High-quality Text-guided 3D Animatable Avatar Generation

TL;DR

X-Oscar is a progressive, step-by-step framework for generating high-quality, animatable 3D avatars from text prompts, addressing oversaturation and quality issues with novel adaptive and score distillation techniques.

Contribution

It introduces a sequential Geometry-Texture-Animation paradigm with Adaptive Variational Parameter and Avatar-aware Score Distillation Sampling for improved avatar generation.

Findings

Outperforms existing text-to-3D and text-to-avatar methods

Produces high-quality, animatable avatars from text prompts

Effectively mitigates oversaturation and low-quality outputs

Abstract

Recent advancements in automatic 3D avatar generation guided by text have made significant progress. However, existing methods have limitations such as oversaturation and low-quality output. To address these challenges, we propose X-Oscar, a progressive framework for generating high-quality animatable avatars from text prompts. It follows a sequential Geometry->Texture->Animation paradigm, simplifying optimization through step-by-step generation. To tackle oversaturation, we introduce Adaptive Variational Parameter (AVP), representing avatars as an adaptive distribution during training. Additionally, we present Avatar-aware Score Distillation Sampling (ASDS), a novel technique that incorporates avatar-aware noise into rendered images for improved generation quality during optimization. Extensive evaluations confirm the superiority of X-Oscar over existing text-to-3D and text-to-avatar approaches. Our anonymous project page: https://xmu-xiaoma666.github.io/Projects/X-Oscar/.