Food image recognition by a replay-based continual learning method using uncertainty-driven past-sample selection

Identifier:  TFM:1885

Handle: https://hdl.handle.net/20.500.11797/TFM1885

Authors:  Pereira Canovas, Anxo-Lois

Abstract:
In this paper, we propose a novel approach to food image recognition utilizing a replay-based continual learning method with uncertainty-driven past-sample selection. Our method aims to address the challenges of data variability and evolving food databases by selectively retaining and revisiting samples based on their uncertain score. The proposed system could improve significantly many industries by improving the benchmarking results of the state-of-the-art methods. We have evaluated our proposed methods and other baseline methods on three datasets, including FOOD101. Finally, we have obtained very positive results, as we have largely outperformed the baseline sample selection methods for rehersal.