CLIP-SLA: Parameter-Efficient CLIP Adaptation for Continuous Sign Language Recognition
Sarah Alyami Hamzah Luqman Information and Computer Science Department, King Fahd University of Petroleum and Minerals Computing Department, Imam Abdulrahman Bin Faisal University SDAIA–KFUPM Joint Research Center for Artificial Intelligence [email protected] [email protected] · 2025
Read CLIP-SLA: Parameter-Efficient CLIP Adaptation for Continuous Sign Language Recognition by Sarah Alyami Hamzah Luqman Information and Computer Science Department, King Fahd University of Petroleum and Minerals Computing Department, Imam Abdulrahman Bin Faisal University SDAIA–KFUPM Joint Research Center for Artificial Intelligence [email protected] [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-01666.txt.
Description
Reading sign language from video means translating a stream of gestures into sentences. Sarah Alyami and Hamzah Luqman borrow the eyes of CLIP, a big vision model trained on images and text. Instead of retraining the whole model, they slip in small trainable pieces. One variant uses adapter modules. The other uses low-rank updates.
Both variants were tested on four sign language datasets, including German and Chinese signing and a new Arabic corpus. Each beat previous state-of-the-art systems while training far fewer parameters. Follow-up experiments showed the approach flexes across different vision-language backbones. The authors pitch it as a cheap, scalable route to sign recognition for real services.
Catalog record
| CREATOR | Sarah Alyami Hamzah Luqman Information and Computer Science Department, King Fahd University of Petroleum and Minerals Computing Department, Imam Abdulrahman Bin Faisal University SDAIA–KFUPM Joint Research Center for Artificial Intelligence [email protected] [email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-01666