# GMGaze: MoE-Based Context-Aware Gaze Estimation with CLIP and Multiscale Transformer

arXiv: https://arxiv.org/abs/2605.00799 · HTML: https://mlsanity.com/paper/2605.00799/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Vision mixture-of-experts, one of 12 directions in Mixture-of-experts.
- Builds on 25 of its 58 references that are in the map.
- First posted on arXiv 1 May 2026.

## Placed in

- Mixture-of-experts / [Vision mixture-of-experts](https://mlsanity.com/moe/vision-mixture-of-experts/)

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

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Domain-Adversarial Training of Neural Networks](https://mlsanity.com/paper/1505.07818/) (2015)
- [Learning to Prompt for Vision-Language Models](https://mlsanity.com/paper/2109.01134/) (2021)
- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [Gaze360: Physically Unconstrained Gaze Estimation in the Wild](https://mlsanity.com/paper/1910.10088/) (2019)
- [ETH-XGaze: A Large Scale Dataset for Gaze Estimation under Extreme Head Pose and Gaze Variation](https://mlsanity.com/paper/2007.15837/) (2020)
- [It's Written All Over Your Face: Full-Face Appearance-Based Gaze Estimation](https://mlsanity.com/paper/1611.08860/) (2016)
- [MPIIGaze: Real-World Dataset and Deep Appearance-Based Gaze Estimation](https://mlsanity.com/paper/1711.09017/) (2017)
- [Appearance-based Gaze Estimation With Deep Learning: A Review and Benchmark](https://mlsanity.com/paper/2104.12668/) (2021)
- [Gaze Estimation using Transformer](https://mlsanity.com/paper/2105.14424/) (2021)
- [A Coarse-to-Fine Adaptive Network for Appearance-Based Gaze Estimation](https://mlsanity.com/paper/2001.00187/) (2020)
- [PureGaze: Purifying Gaze Feature for Generalizable Gaze Estimation](https://mlsanity.com/paper/2103.13173/) (2021)
- [In the Eye of Transformer: Global-Local Correlation for Egocentric Gaze Estimation](https://mlsanity.com/paper/2208.04464/) (2022)
- [Appearance-Based Gaze Estimation Using Dilated-Convolutions](https://mlsanity.com/paper/1903.07296/) (2019)
- [Generalizing Gaze Estimation with Outlier-guided Collaborative Adaptation](https://mlsanity.com/paper/2107.13780/) (2021)
- [Adaptive Feature Fusion Network for Gaze Tracking in Mobile Tablets](https://mlsanity.com/paper/2103.11119/) (2021)
- [DVGaze: Dual-View Gaze Estimation](https://mlsanity.com/paper/2308.10310/) (2023)
- [LatentGaze: Cross-Domain Gaze Estimation through Gaze-Aware Analytic Latent Code Manipulation](https://mlsanity.com/paper/2209.10171/) (2022)
- [Jitter Does Matter: Adapting Gaze Estimation to New Domains](https://mlsanity.com/paper/2210.02082/) (2022)
- [OmniGaze: Reward-inspired Generalizable Gaze Estimation In The Wild](https://mlsanity.com/paper/2510.13660/) (2025)
- [Appearance Debiased Gaze Estimation via Stochastic Subject-Wise Adversarial Learning](https://mlsanity.com/paper/2401.13865/) (2024)
- [Gaze Label Alignment: Alleviating Domain Shift for Gaze Estimation](https://mlsanity.com/paper/2412.15601/) (2024)
- [Differential Contrastive Training for Gaze Estimation](https://mlsanity.com/paper/2502.20128/) (2025)
- [SLYKLatent: A Learning Framework for Gaze Estimation Using Deep Facial Feature Learning](https://mlsanity.com/paper/2402.01555/) (2024)
- [GazeFormer-MoE: Context-Aware Gaze Estimation via CLIP and MoE Transformer](https://mlsanity.com/paper/2601.12316/) (2026)

## Abstract (arXiv)

Gaze estimation methods commonly use facial appearances to predict the direction of a person gaze. However, previous studies show three major challenges with convolutional neural network (CNN)-based, transformer-based, and contrastive language-image pre-training (CLIP)-based methods, including late fusion of image features, lack of factor-aware conditioning, and impractical capacity scaling. To address these challenges, we propose Globally-conditioned Multi-scale Gaze estimation (GMGaze), which leverages a multi-scale transformer architecture. Specifically, the model first introduces semantic prototype conditioning, which modulates the CLIP global image embedding using four learned prototype banks (i.e., illumination, background, head pose and appearance) to generate two complementary context-biased global tokens. These tokens, along with the CLIP patch and CNN tokens, are fused at the first layer. This early unified fusion prevents information loss common in late-stage merging. Finally, each token passes through sparse Mixture-of-Experts modules, providing conditional computational capacity without uniformly increasing dense parameters. For cross-domain adaptation, we incorporate an adversarial domain adaptation technique with a feature separation loss that encourages the two global tokens to remain de-correlated. Experiments using four public benchmarks (MPIIFaceGaze, EYEDIAP, Gaze360, and ETH-XGaze) show that GMGaze achieves mean angular errors of 2.49^circ, 3.22^circ, 10.16^circ, and 1.44^circ, respectively, outperforming previous baselines in all within-domain settings. In cross-domain evaluations, it provides state-of-the-art (SOTA) results on two standard transfer routes.