← Back to all news
arXiv Computer Vision

DiTGaze: Geometry-Conditioned Diffusion Augmentation for Few-Shot Gaze Estimation

DiTGaze is a calibration‑efficient augmentation system for few‑shot gaze estimation that uses a two‑stream 3D Gaussian Splatting backbone and a geometry‑conditioned latent Diffusion Transformer. It bridges rasterized Gaussian features to latent tokens via a learnable projection and frozen VAE, and incorporates timestep, identity, and expression conditioning through Adaptive Layer Normalization. The method densifies gaze supervision with intermediate yaw‑pitch sampling and applies an orthogonality constraint to reduce interference among gaze, pose, and expression, achieving a 4.1 % reduction in generator‑level gaze error on ETH‑XGaze and outperforming baselines across all calibration budgets up to nine real images.

Related stories

arXiv Computer Vision
Sep 17

GazeDiT: Gaze-Accurate Diffusion Image Generation for Eye Tracking via Spatial Conditioning

GazeDiT is a diffusion model that generates synthetic eye images conditioned on a precise 4‑dimensional gaze vector by using a spatial condition derived from pupil and iris geometry. During training, a frozen SegFormer extracts this geometry from real images, while inference employs a physical eye renderer to produce gaze‑consistent geometries without needing a source image. The model achieves lower tail gaze‑label error than other diffusion baselines and improves downstream eye‑tracking accuracy, reducing error on challenging cases from 3.05° to 2.80°.

By Dongze Wu, David Colmenares, Fengting Yang, Jogendra Nath Kundu, Yao Xie, Ali Behrooz, Conny Lu
diffusion
More like this →
Hugging Face Trending Papers
Aug 11

Gaze Target Estimation Anywhere with Concepts

Estimating human gaze targets from images in-the-wild is an important and formidable task. Existing approaches primarily employ brittle, multi-stage pipelines that require explicit inputs, like head bounding boxes and human pose, in order to identify the subject of gaze analysis.

llmsbenchmarks
More like this →
arXiv Computer Vision
Sep 7

EyeMakeYou: Identity-, Task-, and Subjective-State-Conditioned Diffusion for High-Frequency Gaze Synthesis

EyeMakeYou is a multi‑conditional denoising diffusion model that synthesizes high‑frequency, subject‑specific gaze velocity sequences. It conditions on identity, task, and self‑reported subjective states (difficulty, mental tiredness, eye tiredness) to generate realistic 5‑second, 1000‑Hz bivariate gaze data from a reference trajectory. Experiments on the GazeBase dataset show that EyeMakeYou outperforms existing generative methods in spatial accuracy and real‑synthetic similarity while preserving task‑dependent associations with subjective reports.

By Kamrul Hasan, Mehedi Hasan Raju, Oleg V. Komogortsev
ragdiffusion
More like this →
arXiv Computer Vision
Oct 1

GazeFlow: From Human Gaze Behavior to Generative Egocentric Gaze Prediction

GazeFlow is a new framework for egocentric gaze prediction that models gaze as a joint distribution of temporal positions conditioned on both top‑down task cues and bottom‑up visual saliency. It employs conditional flow matching to iteratively transform Gaussian noise into realistic gaze trajectories, using a velocity field informed by video‑encoded visual features and global task queries. On standard benchmarks, GazeFlow outperforms existing methods on per‑frame metrics and produces trajectories that better reflect human gaze dynamics.

By Sheng Zhao, Weikai Lin, Yuhao Zhu
diffusionbenchmarks
More like this →
arXiv Computer Vision
Oct 19

GazeCLIP: Gaze-Guided CLIP with Adaptive-Enhanced Fine-Grained Language Prompt for Deepfake Attribution and Detection

The paper introduces GazeCLIP, a gaze‑guided CLIP framework that uses adaptive‑enhanced fine‑grained language prompts for deepfake attribution and detection (DFAD). It presents a new benchmark evaluating DFAD on novel generative methods such as diffusion and flow models, and proposes a gaze‑aware image encoder (GIE) that fuses gaze prompts with image embeddings to capture global forgery patterns. A language refinement encoder (LRE) further refines language embeddings for precise vision‑language matching, achieving state‑of‑the‑art performance with 6.56% higher accuracy and 5.32% higher AUC on the benchmark.

By Yaning Zhang, Linlin Shen, Zitong Yu, Chunjie Ma, Zan Gao
ragdiffusionmultimodalbenchmarks
More like this →