# CAG-Avatar: Cross-Attention Guided Gaussian Avatars for High-Fidelity Head Reconstruction

arXiv: https://arxiv.org/abs/2601.14844 · HTML: https://mlsanity.com/paper/2601.14844/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 37 references that are in the map.
- First posted on arXiv 21 Jan 2026.

## Builds on (most cited first, 25 of 25)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [The Unreasonable Effectiveness of Deep Features as a Perceptual Metric](https://mlsanity.com/paper/1801.03924/) (2018)
- [3D Gaussian Splatting for Real-Time Radiance Field Rendering](https://mlsanity.com/paper/2308.04079/) (2023)
- [Nerfies: Deformable Neural Radiance Fields](https://mlsanity.com/paper/2011.12948/) (2020)
- [4D Gaussian Splatting for Real-Time Dynamic Scene Rendering](https://mlsanity.com/paper/2310.08528/) (2023)
- [Accelerating 3D Deep Learning with PyTorch3D](https://mlsanity.com/paper/2007.08501/) (2020)
- [Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction](https://mlsanity.com/paper/2309.13101/) (2023)
- [Dynamic Neural Radiance Fields for Monocular 4D Facial Avatar Reconstruction](https://mlsanity.com/paper/2012.03065/) (2020)
- [AD-NeRF: Audio Driven Neural Radiance Fields for Talking Head Synthesis](https://mlsanity.com/paper/2103.11078/) (2021)
- [BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation](https://mlsanity.com/paper/2004.02147/) (2020)
- [HeadNeRF: A Real-time NeRF-based Parametric Head Model](https://mlsanity.com/paper/2112.05637/) (2021)
- [SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting](https://mlsanity.com/paper/2403.05087/) (2024)
- [Towards Metrical Reconstruction of Human Faces](https://mlsanity.com/paper/2204.06607/) (2022)
- [Instant Volumetric Head Avatars](https://mlsanity.com/paper/2211.12499/) (2022)
- [Robust High-Resolution Video Matting with Temporal Guidance](https://mlsanity.com/paper/2108.11515/) (2021)
- [RigNeRF: Fully Controllable Neural 3D Portraits](https://mlsanity.com/paper/2206.06481/) (2022)
- [FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding](https://mlsanity.com/paper/2312.02214/) (2023)
- [MonoGaussianAvatar: Monocular Gaussian Point-based Head Avatar](https://mlsanity.com/paper/2312.04558/) (2023)
- [KeypointNeRF: Generalizing Image-based Volumetric Avatars using Relative Spatial Encoding of Keypoints](https://mlsanity.com/paper/2205.04992/) (2022)
- [AvatarMAV: Fast 3D Head Avatar Reconstruction Using Motion-Aware Neural Voxels](https://mlsanity.com/paper/2211.13206/) (2022)
- [Learning Personalized High Quality Volumetric Head Avatars from Monocular RGB Videos](https://mlsanity.com/paper/2304.01436/) (2023)
- [HACK: Learning a Parametric Head and Neck Model for High-fidelity Animation](https://mlsanity.com/paper/2305.04469/) (2023)

## Abstract (arXiv)

Creating high-fidelity, real-time drivable 3D head avatars is a core challenge in digital animation. While 3D Gaussian Splashing (3D-GS) offers unprecedented rendering speed and quality, current animation techniques often rely on a "one-size-fits-all" global tuning approach, where all Gaussian primitives are uniformly driven by a single expression code. This simplistic approach fails to unravel the distinct dynamics of different facial regions, such as deformable skin versus rigid teeth, leading to significant blurring and distortion artifacts. We introduce Conditionally-Adaptive Gaussian Avatars (CAG-Avatar), a framework that resolves this key limitation. At its core is a Conditionally Adaptive Fusion Module built on cross-attention. This mechanism empowers each 3D Gaussian to act as a query, adaptively extracting relevant driving signals from the global expression code based on its canonical position. This "tailor-made" conditioning strategy drastically enhances the modeling of fine-grained, localized dynamics. Our experiments confirm a significant improvement in reconstruction fidelity, particularly for challenging regions such as teeth, while preserving real-time rendering performance.