no code implementations • 26 Nov 2023 • Jan Olszewski, Dawid Rymarczyk, Piotr Wójcik, Mateusz Pach, Bartosz Zieliński
To reduce this computational inefficiency, we introduce the TOken REcycling (TORE) modification for the ViT inference, which can be used with any architecture.
no code implementations • 29 Jun 2023 • Piotr Wójcik, Hussein Naji, Adrian Simon, Reinhard Büttner, Katarzyna Bożek
Inspired by Bidirectional Encoder representation from Image Transformers (BEiT), we split the images into smaller patches and generate corresponding discrete visual tokens.