Efficient 2-D/3-D Gaze Estimation Using TGGNet: A Transformer Graph Approach
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The human eye gaze is a crucial visual and cognitive attention indicator, with broad applications in intelligent vehicle systems and human-machine interaction. This paper presents a novel gaze estimation approach using Graph Neural Networks (GNNs), leveraging the geometric relationship between facial landmarks and gaze direction. Existing appearance-based gaze estimation approaches primarily rely on raw facial images, often overlooking the spatial relationships between facial landmarks and gaze direction. Additionally, many recent methods involve large, computationally expensive models, limiting their applicability in real-time scenarios. Facial landmarks serve as graph nodes, and spatial distances form the edges. We demonstrate significant correlations between node positions and gaze direction, as well as between edge lengths and head pose. Our Transformer Graph Gaze Network (TGGNet) processes this graph-based data to estimate the gaze direction. The lightweight Transformer-based GNN model, with approximately 3.72 million parameters and only 0.76 Giga FLOPs, is highly suitable for real-time systems, offering both computational efficiency and low memory requirements. TGGNet assigns higher attention weights to key landmarks, improving gaze estimation. We validated the model on GazeCapture and MPIIFaceGaze (2D) and Gaze360 (3D), showing superior performance. Attention map analysis highlights the importance of landmarks around the eyes, particularly the pupils, irises, and eyelids. Video demos and codes can be found on our project’s repository <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/AiX-Lab-UWO/GazeTGGNet</uri>.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it