Amrita SLR- Real Time ISL Dataset

The Amrita SLR dataset has been developed to evaluate sign language recognition models under realistic conditions, with a focus on real-time applicability. It contains video recordings of 50 commonly used Indian Sign Language (ISL) signs, captured across diverse environmental settings and signer variations.

Each video is recorded at a resolution of 1920×1080 pixels with a frame rate of 28 FPS, resulting in an average of 56 frames per video and an average video duration of 2.57 seconds. The dataset includes approximately 16 samples per class, collected from 7 different signers representing a range of age groups and signing styles.

To simulate practical usage conditions, videos were recorded using seven different mobile camera devices with resolutions ranging from 16 MP to 50 MP. The dataset features significant background variability, including both indoor and outdoor environments. Furthermore, distance variation between the signer and the recording device was intentionally included to capture realistic signing scenarios.
The signers were recorded in both standing and sitting positions, introducing additional variation in hand movement dynamics and body posture. These attributes, along with the inclusion of signing style diversity, make the dataset suitable for evaluating the robustness and generalizability of sign language recognition systems, especially in real-time deployment contexts.