aiqbench analysis
This research introduces TransSLR, a lightweight temporal transformer that significantly improves sign language recognition accuracy for underrepresented languages like Central African Sign Language by utilizing geometric keypoint data instead of raw video frames. The model achieves an 80.39% top-1 accuracy on the CASL-W60 benchmark, demonstrating a substantial leap over prior state-of-the-art results while maintaining a manageable parameter count of approximately 8.7 million. This work highlights critical advancements in making AI-driven accessibility tools viable for low-resource linguistic communities through efficient architecture design and pose-based representation learning.
TransSLR: A Lightweight Transformer for Sign Language Recognition
arXiv:2608.06407v2 Announce Type: replace-cross Abstract: Automated sign language recognition for underrepresented languages remains a largely unsolved problem. Central African Sign Language (CASL) exemplifies this challenge: the CASL-W60 benchmark contains 60 isolated signs, and the best published result is 69.93%. We present TransSLR, a lightweight Temporal Transformer Encoder trained from scratch on normalized 64-frame pose sequences. By operating on geometric keypoint representations rather than raw RGB, TransSLR achieves 80.39% top-1 accuracy and 91.07% top-5 accuracy on a signer-independent evaluation split. This result is 10.46 percentage points higher than the published 69.93% result. TransSLR has 8.67 million trainable parameters. We also evaluate zero-shot transfer from a high-resource sign-language model, which achieves 0.00% exact-match accuracy under our manual gloss-matching protocol. Our experiments compare pose-based, RGB-based, and multimodal approaches to sign-language recognition for a low-resource language.
Abstract:Automated sign language recognition for underrepresented languages remains a largely unsolved problem. Central African Sign Language (CASL) exemplifies this challenge: the CASL-W60 benchmark contains 60 isolated signs, and the best published result is 69.93%. We present TransSLR, a lightweight Temporal Transformer Encoder trained from scratch on normalized 64-frame pose sequences. By operating on geometric keypoint representations rather than raw RGB, TransSLR achieves 80.39% top-1 accuracy and 91.07% top-5 accuracy on a signer-independent evaluation split. This result is 10.46 percentage points higher than the published 69.93% result. TransSLR has 8.67 million trainable parameters. We also evaluate zero-shot transfer from a high-resource sign-language model, which achieves 0.00% exact-match accuracy under our manual gloss-matching protocol. Our experiments compare pose-based, RGB-based, and multimodal approaches to sign-language recognition for a low-resource language.
This paper has been accepted for oral presentation at Deep Learning Indaba 2026 which will be hosted on the IJCAI hosting platform. paper link: this https URL
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2608.06407 [cs.CV]
(or arXiv:2608.06407v2 [cs.CV] for this version)
Full text under the publisher's open licence · 171 words
Related coverage
A Low-Cost Hybrid Reservoir Computing Model for Isolated Sign Language Video Recognition
arXiv · Aug 5, 2026
SignBind-LLM: Multi-Stage Modality Fusion for Sign Language Translation
arXiv · Sep 3, 2026
SignDino: Self-Supervised Sign Language Representation Learning via Temporal-Axis Self-Distillation
arXiv · Sep 9, 2026
PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters
Hugging Face · Jun 22, 2026