Hugo André Sousa Gomes

Mr. Hugo André Sousa Gomes

System's and Informatic's engineering department, Polytechnic Institute of Coimbra, Coimbra, Portugal
orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 November 2022 in Energies
Verified
Journal Article
Open access
PDF file
Hugo André Sousa GomesHugo André Sousa Gomes
Hugo André Sousa Gomes
NRNuno Redinha
Nuno Redinha
Nuno LavadoNuno Lavado
Nuno Lavado
mateus mendesmateus mendes
mateus mendes
Counting objects in video images has been an active area of computer vision for decades. For precise counting, it is necessary to detect objects and follow them through consecutive frames. Deep neural networks have allowed great improvements in this area. Nonetheless, this task is still a challenge for edge computing, especially when low-power edge AI devices must be used. The present work describes an application where an edge device is used to run a YOLO network and V-IOU tracker to count people and bicycles in real time. A selective frame-downsampling algorithm is used to allow a larger frame rate when necessary while optimizing memory usage and energy consumption. In the experiments, the system was able to detect and count the objects with 18 counting errors in 525 objects and a mean inference time of 112.82 ms per frame. With the selective downsampling algorithm, it was also capable of recovering and reduce memory usage while maintaining its precision.

25 November 2019 in arXiv
Verified
Preprint
Open access
PDF file
AMAzadeh Sadat Mozafari
Azadeh Sadat Mozafari
Hugo André Sousa GomesHugo André Sousa Gomes
Hugo André Sousa Gomes
CGChristian Gagné
Christian Gagné
The uncertainty estimation is critical in real-world decision making applications, especially when distributional shift between the training and test data are prevalent. Many calibration methods in the literature have been proposed to improve the predictive uncertainty of DNNs which are generally not well-calibrated. However, none of them is specifically designed to work properly under domain shift condition. In this paper, we propose Unsupervised Temperature Scaling (UTS) as a robust calibration method to domain shift. It exploits unlabeled test samples instead of the training one to adjust the uncertainty prediction of deep models towards the test distribution. UTS utilizes a novel loss function, weighted NLL, which allows unsupervised calibration. We evaluate UTS on a wide range of model-datasets to show the possibility of calibration without labels and demonstrate the robustness of UTS compared to other methods (e.g., TS, MC-dropout, SVI, ensembles) in shifted domains.