Пошук академічних текстів

Інформація × Реєстраційний номер 2123U011183, Матеріали видань та локальних репозитаріїв Категорія Препринт Назва роботи Research of Data Augmentation Approaches for Enhancing Classification Model Performance Автор Vey BohdanVey Bohdan Дата публікації 01-01-2023 Постачальник інформації Український католицький університет Першоджерело https://hdl.handle.net/20.500.14570/4385 Видання Опис After a significant improvement in the computational powers of modern comput- ers, the models became larger, and their accuracy increased. However, due to a high amount of parameters, modern neural networks also need much bigger datasets for efficient usage. Augmentation partly solves this problem, but the most up-to- date augmentation still doesn’t change the image patterns. We propose a new way of augmentation by using inpainting models to change the image’s nature. Then we compare model performance by using traditional augmentation and GANAug- mentation. The second part of this study will use Test Time Augmentation(TTA) to improve model performance for data which come from another source. Додано в НРАТ 2025-11-05 Закрити