Semi-Supervised Learning for Hand Gesture Recognition
Course: CS7387 – Research in Computer Science
Presenter: Parisa Tabassum
Advisor: Dr. Vangelis Metsis
Date/Time: May 11, 11:00 a.m.
Location: https://txstate.zoom.us/j/97183479778
Abstract:
Sign language is a form of nonverbal communication in which language is communicated by hand gestures. The ability to distinguish between human gestures can help machines perform better in a variety of applications, including sign language recognition, assisted living, and healthcare. To this end, this study aims to improve accuracy of hand gesture recognition using a state-of-the-art semi-supervised learning strategy. Semi-supervised learning is a machine learning technique that integrates a small amount of labeled data with a large amount of unlabeled data during training. It employs self-supervised contrastive representation learning from unlabeled training data with the goal of learning an embedding space in which similar sample pairs (positive pair) stay close to each other while dissimilar ones (negative pair) are pushed apart. This study utilizes nearest neighbor contrastive learning of visual representations (NNCLR) as a self-supervised learning approach. This technique finds positives from other instances of the dataset by treating nearest neighbors in the learned representation space as positives. The network is then fine-tuned using a dataset with only a subset of the labels present. The 'Hand Gesture Recognition Database,' which consists of 10 different hand-gestures performed by 10 different participants, was used to evaluate the performance of this methodology.
Deadline: June 8, 2022, midnight