Research map / Video / Video segmentation and tracking
Fingerspelling Recognition in the Wild With Iterative Visual Attention
Bowen Shi, Aurora Martinez Del Rio, Jonathan Keane, Diane Brentari, Greg Shakhnarovich, Karen Livescu
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 standard visual attention (ablation) | whole frame · ChicagoFSWild dev | letter accuracy (%): 11.0 -> 23.0 | Improved |
| replace standard attention with iterative attention (ablation) | whole frame+attention · ChicagoFSWild dev | letter accuracy (%): 23.0 -> 42.8 | Improved |
| replace standard attention with iterative attention (ablation) | face ROI+attention · ChicagoFSWild dev | letter accuracy (%): 33.4 -> 45.6 | Improved |
| replace standard attention with iterative attention (ablation) | hand ROI+attention · ChicagoFSWild dev | letter accuracy (%): 41.4 -> 45.0 | Improved |
| add language model (ablation) | Ours+face ROI · ChicagoFSWild dev | letter accuracy (%): 45.6 -> 46.8 | Improved |
| add language model (ablation) | Ours+whole frame · ChicagoFSWild dev | letter accuracy (%): 42.8 -> 43.6 | Improved |
| add ChicagoFSWild+ training data (ablation) | Ours+face ROI+LM · ChicagoFSWild test | letter accuracy (%): 45.1 -> 61.2 | Improved |
| add ChicagoFSWild+ training data (ablation) | Hand ROI+LM · ChicagoFSWild test | letter accuracy (%): 41.9 -> 57.5 | Improved |
| add ChicagoFSWild+ training data (ablation) | Ours+whole frame+LM · ChicagoFSWild+ test | letter accuracy (%): 43.8 -> 61.0 | Improved |
| zoom into region instead of enlarging input, with motion prior (ablation) | Ours+face ROI · ChicagoFSWild dev, face ROI, R=0.9^3 | letter accuracy (%): 39.3 -> 39.6 | No clear effect |
| zoom into region instead of enlarging input, no motion prior (ablation) | Ours+face ROI · ChicagoFSWild dev, face ROI, R=0.9^3 | letter accuracy (%): 38.1 -> 39.8 | Improved |
| use iterative-attention ROI as hand detector instead of off-the-shelf detector (ablation) | face ROI model · ChicagoFSWild hand annotations (200 frames) | average IoU at miss rate 0.158: 0.213 -> 0.443 | Improved |
Closest papers by meaning
- Fingerspelling Detection in American Sign Language (CVPR 2021 · similarity 0.92)
- FSboard: Over 3 Million Characters of ASL Fingerspelling Collected via Smartphones (CVPR 2025 · similarity 0.91)
- OpenFS: Multi-Hand-Capable Fingerspelling Recognition with Implicit Signing-Hand Detection and Frame-Wise Letter-Conditioned Synthesis (CVPR 2026 · similarity 0.90)
- BANZ-FS: BANZSL Fingerspelling Dataset (ICLR 2026 · similarity 0.89)
- A Wireframe-Based Approach for Classifying and Acquiring Proficiency in the American Sign Language (Student Abstract) (AAAI 2024 · similarity 0.87)
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.85)
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.
