Research map / Architectures and efficiency / Pruning and efficient tuning
Alfa: Attentive Low-Rank Filter Adaptation for Structure-Aware Cross-Domain Personalized Gaze Estimation
He-Yen Hsieh, Wei-Te Mark Ting, H. T. Kung
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 |
|---|---|---|---|
| add attention over rank slices (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 7.29 -> 5.72 | Improved |
| increase attention heads from 1 to 16 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 6.20 -> 5.72 | Improved |
| increase attention heads from 16 to 32 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 5.72 -> 5.91 | Hurt |
| use LoRA rank 8 instead of 4 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 5.84 -> 5.72 | Improved |
| use LoRA rank 16 instead of 8 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 5.72 -> 5.90 | Hurt |
| use SVD rank 64 instead of 32 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 6.69 -> 5.72 | Improved |
| use SVD rank 64 instead of 128 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 5.90 -> 5.72 | Improved |
| use 5 personal samples instead of 1 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 6.13 -> 5.72 | Improved |
| use 10 personal samples instead of 5 (ablation) | Alfa · Gaze cross-domain avg | angular gaze error (deg): 5.72 -> 5.70 | No clear effect |
| use Alfa instead of LoRA with same SVD-truncated weights | gaze model (SVD rank 64) · Gaze cross-domain avg | angular gaze error (deg): 5.95 -> 5.72 | Improved |
| use Alfa instead of LoRA rank 128 on LLaDA-8B-Instruct | LLaDA-8B-Instruct · Countdown (0-shot, 128) | accuracy: 20.3 -> 27.3 | Improved |
| use Alfa instead of LoRA rank 128 on LLaDA-8B-Instruct | LLaDA-8B-Instruct · Sudoku (0-shot, 128) | accuracy: 16.5 -> 9.7 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Few-Shot Adaptive Gaze Estimation (ICCV 2019 · similarity 0.91)
- Render-to-Adapt: Unsupervised Personal Adaptation for Gaze Estimation (CVPR 2026 · similarity 0.91)
- Test-Time Personalization with Meta Prompt for Gaze Estimation (AAAI 2024 · similarity 0.89)
- Gaze from Origin: Learning for Generalized Gaze Estimation by Embedding the Gaze Frontalization Process (AAAI 2024 · similarity 0.89)
- CLIP-Gaze: Towards General Gaze Estimation via Visual-Linguistic Model (AAAI 2024 · similarity 0.88)
- Bootstrap AutoEncoders With Contrastive Paradigm for Self-supervised Gaze Estimation (ICML 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.
