Research map / Image generation / Image restoration
A Differential Approach for Gaze Estimation
Gang Liu, Yu Yu, Kenneth Alberto Funes Mora, Jean‐Marc Odobez
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| predict gaze differences between image pairs instead of absolute gaze, with linear calibration | baseline CNN · MPIIGaze | average angular error (deg): 5.67 -> 4.64 | Improved |
| predict gaze differences instead of absolute gaze | Lin-Ad · EYEDIAP | average angular error (deg): 3.84 -> 3.23 | Improved |
| predict gaze differences instead of absolute gaze | Lin-Ad · UT-Multiview | average angular error (deg): 4.56 -> 4.13 | Improved |
| add Gaussian weighting of per-reference predictions (ablation) | Diff-NN · EYEDIAP | average angular error (deg): 3.36 -> 3.23 | Improved |
| fine-tune differential network on reference pairs (Diff-NN-Ad) | Diff-NN · EYEDIAP | average angular error (deg): 3.23 -> 3.00 | Improved |
| fine-tune differential network on reference pairs (Diff-NN-Ad) | Diff-NN · MPIIGaze | average angular error (deg): 4.64 -> 4.59 | No clear effect |
| replace C1-C3 blocks with VGG-16 backbone | Diff-NN · MPIIGaze | average angular error (deg): 4.64 -> 3.80 | Improved |
| use mid-level fusion instead of early fusion (ablation) | differential network · EYEDIAP | average angular error (deg): 3.40 -> 3.23 | Improved |
| use mid-level fusion instead of late fusion (ablation) | differential network · EYEDIAP | average angular error (deg): 3.47 -> 3.23 | Improved |
| train multi-task absolute and difference gaze with Lin-Ad instead of difference only (ablation) | siamese network · EYEDIAP | average angular error (deg): 3.40 -> 3.76 | Hurt |
| vary Gaussian bandwidth sigma from 5.7 to 22.8 degrees (ablation) | Diff-NN · EYEDIAP | average angular error (deg): changes in a narrow range of 0.1 | No clear effect |
| draw references from a single day and test on other days | Diff-NN · MPIIGaze | average angular error (deg): 4.64 -> 4.83 | Hurt |
Closest papers by meaning
- Towards High Performance Low Complexity Calibration in Appearance Based Gaze Estimation (TPAMI 2022 · similarity 0.90)
- Deep Pictorial Gaze Estimation (ECCV 2018 · similarity 0.89)
- Few-Shot Adaptive Gaze Estimation (ICCV 2019 · similarity 0.89)
- A Coarse-to-Fine Adaptive Network for Appearance-Based Gaze Estimation (AAAI 2020 · similarity 0.89)
- Improving Few-Shot User-Specific Gaze Adaptation via Gaze Redirection Synthesis (CVPR 2019 · similarity 0.89)
- Gaze from Origin: Learning for Generalized Gaze Estimation by Embedding the Gaze Frontalization Process (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
