Table 6 Comparison with recent state-of-the-art ISL recognition models.
Model | Dataset | Accuracy / mAP | FPS | Year |
|---|---|---|---|---|
CNN + LSTM36 | ISLRTC | 91.2% | 16.5 | 2020 |
3D CNN42 | Custom | 92.5% | 12.3 | 2024 |
Transformer-based41 | MultiBench | 92.8% | 14.0 | 2021 |
Faster R-CNN63 | Custom | 93.1 | 22.3 | 2022 |
SSD64 | Custom | 93.7 | 28.5 | 2025 |
RetinaNet65 | Custom | 94.5 | 35.9 | 2025 |
YOLOv544 | Custom | 94.8% | 36.1 | 2024 |
YOLOv845 | Custom | 95.2% | 39.2 | 2024 |
YOLOv10-ST (Proposed) | Our Dataset | 97.62% | 48.7 | 2025 |
