ISL Dataset (Dynamic and Static)

The sign dataset consists of 26 classes (if “Where” is also considered) and were made with the reference of Sign Dictionary released by Indian Sign Language Research and Training Centre (ISLRTC). We ensured that compound signs were broken down into their constituent meaningful signs, e.g. “Where” is made up of 2 signs, “Place” and “What”. Thus, separate classes were made for these signs as this provides us the flexibility for combining signs to form different phrases and sentences, which is an essential part of any natural language with its grammar and syntax. MediaPipe was used to create the dataset ensuring different camera angles and covered variations of each sign. Each class contained 100 elements and within each recording, 30 frames were extracted and stored as a NumPy array. The NumPy arrays store the key-points extracted from each frame i.e. the Pose Contours and Hand Connections.