Tools
Tools
Das, Semanta (2025) Word-Based American Sign Language Recognition Using 2D CNN with Temporal Shift Module and Attention. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (817kB) | Preview |
Abstract
This work presents a lightweight American Sign Language (ASL) word recognition system built on a 2D CNN backbone (MobileNet-V2) enhanced with Temporal Attention and a Temporal Shift Module (TSM). The system was trained and evaluated on the 100-word subset of the MS-ASL dataset. A four-stage incremental development process was followed, where the weights from each stage were transferred to initialize the next which enables progressive learning. The process began with a frame-level pretraining in Stage A. From Stage B, temporal attention was applied, which helped to get better accuracy. In Stage C, stronger data augmentation was applied, with a dropout rate of 0.5 which helped reduce overfitting. Finally, in the Stage D, TSM was introduced with attention which enhanced temporal modelling. The final model achieved a top-1 accuracy of 51.97% on the MS-ASL100 dataset. Compared with model like 3D CNNs and Transformer, my proposed model is not that much effective, as we are only working with 2D kernel.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Tomer, Vikas UNSPECIFIED |
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science H Social Sciences > HM Sociology > Information Science > Communication P Language and Literature > P Philology. Linguistics > Semiotics > Language. Linguistic theory > Gesture. Sign language |
| Divisions: | School of Computing > Master of Science in Data Analytics |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 25 Aug 2026 11:44 |
| Last Modified: | 25 Aug 2026 11:44 |
| URI: | https://norma.ncirl.ie/id/eprint/9625 |
Actions (login required)
 |
View Item |