Using the landmarks coordinates for hand gestures {x,y,z} generated using mediapipe i transformed the dataset to a graph classification problem. Our model also achieved quite a good accuracy - 90% on training set and 86% on test set.
rimo02/Sign-Language-Detection
Detecting American Sign Language using Graph Conv Network and Mediapipe
PythonMIT