Publications
In this section I show my most recent published papers. The complete list can be found on my Google Scholar profile.
A. Menezes, Gustavo de Moura, C. Alves, André C. P. L. F. de Carvalho.
ArXiv - May 2022
This systematic review paper explores the field of Continual Learning, primarily focusing on class-incremental object detection. In this review, we analyze the current strategies proposed to tackle the problem of class-incremental object detection. We provide a short and systematic review of the methods, an evaluation of the existing approaches, and an overview of the current trends and possible future research directions.
Available on arXiv.
Submitted to the Neural Networks Journal.
Gustavo de Moura Souza, Claudio F. M. Toledo.
WCCI CEC - July 2020
The present paper introduces a hybrid genetic algorithm for path planning problem with obstacle avoidance. The genetic algorithm is combined with Ray Casting (RC) algorithm, where RC is responsible to avoid obstacles and to find safe regions for emergency landing. Thus, the path planning system must deal with a non-convex environment when planning and re-planning trajectories. The system must also work embedded on the UAV running under a Raspberry Pi board. The hybrid method is evaluated over 50 benchmark maps from literature with satisfactory results reported.
Presented on the WCCI Congress on Evolutionary Computation (CEC) 2020.
M. Carvalho*, A. Alcântara*, Gustavo de Moura Souza*, K. Bogdan*, M. Martins, E. Gadbem, L. Maia.
Rio Oil & Gas - December 2020
In this work, we present an approach based on Optical Character Recognition (OCR) and text processing techniques to automatically extract information from digital well drilling schematics reports to a structured template. We present results that indicates this work can be a starting point for the extraction of structured information from unstructured/semi-structured reports in the O&G field.
Presented on the Rio Oil & Gas 2020 conference.
*equal contribution
K. Bogdan, G. Megeto, R. Leal, Gustavo Souza, A. Valente, L. Kirsten .
ICDAR - September 2021
In this work, we propose a lightweight neural network to enhance photographed document images using a feature extraction that aggregates multi-scale features and a pixel-wise adjustment refinement step. We focused on aesthetics aspects of the images. Our method was able to lessen the negative effects of different artifacts, such as shadows and insufficient illumination, while also maintaining a good color consistency, resulting in a better final enhanced image.
Presented on ICDAR 2021 conference.