Learning-by-Synthesis for Appearance-Based 3D Gaze Estimation

By Yusuke Sugano, Yasuyuki Matsushita, Yoichi Sato. 2014

424 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Ophthalmology, Radiology, Nuclear Medicine and Imaging, 3D Vision & Pose, Human-Computer Interaction

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Ophthalmology

In Radiology, Nuclear Medicine and Imaging

In 3D Vision & Pose

In Human-Computer Interaction

Citations in the corpus

Seeing with Humans: Gaze-Assisted Neural Image Captioning (2016, background): “Recent advances in crowd-sourcing techniques [6], [7], [8] and appearance-based estimation methods [9], [10], [11] are also further paving the way for low-cost and large-scale gaze data collection.”
ETH-XGaze: A Large Scale Dataset for Gaze Estimation under Extreme Head Pose and Gaze Variation (2020, background): “UT Multi-view (UTMV) [33] is recorded with eight machine version cameras, and HUMBI is recorded with multiple wide-angle cameras, however, their resolution of eye region is small in the captured image.”
Fake it till you make it: face analysis in the wild using synthetic data alone (2021, methodology): “Building parametric models is challenging, so an alternative is to render 3D scans directly [4, 55, 62, 68].”
Controllable Continuous Gaze Redirection (2020, background): “Some publicly available gaze datasets, such as MPIIGaze [43], Gaze Capture [17], UT Multi-View [30] or CMU Multi-Pie [11], only provide low-resolution images and would therefore introduce an unwanted bias towards low-quality images.”
It’s Written All Over Your Face: Full-Face Appearance-Based Gaze Estimation (2016, methodology): “proposed a data normalisation technique to restrict the appearance variation into a single, normalized training space [25].”
Eye Tracking for Everyone (2016, background): “Gaze datasets: There are a number of publicly available gaze datasets in the community [24, 40, 31, 25, 34, 43, 13].”
Rendering of Eyes for Eye-Shape Registration and Gaze Estimation (2015, methodology): “We synthesized training images using the same camera settings as in the UT dataset [9].”
Gaze360: Physically Unconstrained Gaze Estimation in the Wild (2019, background): “TabletGaze [11] 2D ∼80 51 Eyes No iTracker [12] 2D ∼100 1,450 Full Partially UT MV [23] 3D ∼50 50 Eyes No Columbia [20] 3D 60 56 Full* No RT-GENE [4] 3D 75 15 Full* No MPIIFaceGaze [31] 3D ∼80 15 Full No EYEDIAP [17] 3D 90 16 Full No Weidenb.”
RT-GENE: Real-Time Eye Gaze Estimation in Natural Environments (2018, methodology): “Gaze estimation performance comparison: We evaluated our method on two de facto standard datasets, MPII Gaze [60] and UT Multi-view [53](3), as well as our newly proposed RT-GENE dataset.”
MPIIGaze: Real-World Dataset and Deep Appearance-Based Gaze Estimation (2017, methodology): “Table 1): UT Multiview [6] (top), EYEDIAP [12] (middle), and Columbia [13] (bottom).”
Learning an appearance-based gaze estimator from one million synthesised images (2016, methodology): “While some previous datasets captured participants from a sparse and discrete set of different camera angles [Smith et al. 2013; Sugano et al. 2014], we instead specified the transforms of camera and eyeball continuously, allowing us to synthesize dense training data.”