Bringing balance to hand shape classification: Mitigating data imbalance through generative models

Open Webpage

Gaston Gustavo Rios, Pedro Dal Bianco, Franco Ronchetti, Facundo Manuel Quiroga, Oscar Stanchi, Santiago Ponte Ahón, Waldo Hasperué

Published: 01 Jan 2025, Last Modified: 26 Sept 2025Appl. Soft Comput. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Generated data improves handshape classification on limited and unbalanced data.•Pre-training with generated data and fine-tuning with real data boosts performance.•Dataset balance through generative models boosts per-class accuracy by up to 100%.•Models pre-trained with generated samples achieve an earlier convergence.•Our models set a new state-of-the-art for the RWTH handshape dataset.