LSFB-CONT and LSFB-ISOL: Two New Datasets for Vision-Based Sign Language Recognition

While significant progress have been made in thefield of Natural Language Processing (NLP), leading the com-mercially available products, Sign Language Recognition (SLR)is still in its infancy. The lack of large-scale sign languagedatasets makes it hard to leverage new Deep Learning methods.In this paper, we introduce LSFB-CONT, a large scale datasetsuited for continuous SLR along with LSFB-ISOL, a subset ofLSFB-CONT for isolated SLR. Baseline SLR experiments areconducted on LSFB-ISOL and the reached accuracy measuresare compared with those obtained on previous datasets. Theresults suggest that state-of-the-art models for action recognitionstill lack sufficient internal representation power to capture thehigh level of variations of a sign language.