Implementation and evaluation of models for automatic segmentation of continuous sign language into sentence units.
sl-segmentation: Implementation of the segmentation model based in optical flow and experiments.st-gcn: Implementation of a segmentation model that uses spatio-temporal graph convolutional networks for processing pose sequences and experiments.
Real-Time Sign Language Detection using Human Pose Estimation (Code)
A TensorFlow implementation of the model presented in "Real-Time Sign Language Detection using Human Pose Estimation", published at SLRTP 2020.
Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition (Code)
A PyTorch implementation of the model presented in "Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition", adapted for the task of sign language video segmentation, published at AAAI-18. The applied approach is similar to the one used in "Automatic Segmentation of Sign Language into Subtitle-Units" (Code), published at SLRTP 2020.
Extraction of skeletons using Mediapipe and conversion to OpenPose format: Code