Mohsen Saffari

Bio: Mohsen Saffari is an academic researcher from University of Tulsa. The author has contributed to research in topics: Facial recognition system & Computer science. The author has an hindex of 2, co-authored 8 publications receiving 12 citations. Previous affiliations of Mohsen Saffari include K.N.Toosi University of Technology & University of Porto.

MFR 2021: Masked Face Recognition Competition

Abstract: This paper presents a summary of the Masked Face Recognition Competitions (MFR) held within the 2021 International Joint Conference on Biometrics (IJCB 2021). The competition attracted a total of 10 participating teams with valid submissions. The affiliations of these teams are diverse and associated with academia and industry in nine different countries. These teams successfully submitted 18 valid solutions. The competition is designed to motivate solutions aiming at enhancing the face recognition accuracy of masked faces. Moreover, the competition considered the deployability of the proposed solutions by taking the compactness of the face recognition models into account. A private dataset representing a collaborative, multisession, real masked, capture scenario is used to evaluate the submitted solutions. In comparison to one of the topperforming academic face recognition solutions, 10 out of the 18 submitted solutions did score higher masked face verification accuracy.

My Eyes Are Up Here: Promoting Focus on Uncovered Regions in Masked Face Recognition

Abstract: The recent Covid-19 pandemic and the fact that wearing masks in public is now mandatory in several countries, created challenges in the use of face recognition systems (FRS). In this work, we address the challenge of masked face recognition (MFR) and focus on evaluating the verification performance in FRS when verifying masked vs unmasked faces compared to verifying only unmasked faces. We propose a methodology that combines the traditional triplet loss and the mean squared error (MSE) intending to improve the robustness of an MFR system in the masked-unmasked comparison mode. The results obtained by our proposed method show improvements in a detailed step-wise ablation study. The conducted study showed significant performance gains induced by our proposed training paradigm and modified triplet loss on two evaluation databases.

Deep Convolutional Graph Rough Variational Auto-Encoder for Short-Term Photovoltaic Power Forecasting

Abstract: Photovoltaic (PV) power is considered as one of the most promising sustainable energy resources in recent years. However, the existing intermittency in the nature of solar energy is a significant problem for the optimization of smart grids. In this paper, to overcome PV generation uncertainty and provide an accurate spatio-temporal (ST) PV forecast, we propose a novel deep generative convolutional graph rough variational autoencoder (CGRVAE) that captures each PV site's probability distribution functions (PDFs) of future PV generation in a modeled weighted graph. Having the learned PDFs enables CGRVAE to accurately generate the future values of PV power time series. To train and evaluate our model, we used the measurements of a set of PV sites in California, US. The sites are modeled as a weighted graph where each node represents PV measurements at each site while edges reflect their correlations. Using graph spectral convolutions the proposed model extracts the most relevant information of the graph to estimate the future PV given the historical time series for each node in the modeled graph. Experimental results show the superiority of CGRVAE over state-of-the-art forecasting approaches in terms of the root mean square error (RMSE) and mean absolute error (MAE) metric.

Neural Control of Mobile Robot Motion Based on Feedback Error Learning and Mimetic Structure

Abstract: Mobile robots motion control is a basic problem in robotics and there are still some control difficulties such as uncertainty in a real implementation which should be considered. This paper is concerned with the neural control of wheeled mobile robots trajectory tracking and posture stabilization. In the trajectory-tracking problem, the Feedback Error Learning (FEL) structure is used and for the posture stabilization problem, the Mimetic structure is employed. These neural based structures use a classic controller, Dynamic Feedback Linearization (DFL), and help to improve the adaptiveness of it. The effectiveness of the proposed controllers is verified by simulation in Webots robotic simulator and on the e-puck which is a differential wheeled mobile robot. The simulation results verify the ability of the proposed methods for controlling the robot and handling uncertainties.

A Survey on Masked Facial Detection Methods and Datasets for Fighting Against COVID-19

Abstract: Coronavirus disease 2019 (COVID-19) continues to pose a great challenge to the world since its outbreak. To fight against the disease, a series of artificial intelligence (AI) techniques are developed and applied to real-world scenarios such as safety monitoring, disease diagnosis, infection risk assessment, and lesion segmentation of COVID-19 CT scans. The coronavirus epidemics have forced people wear masks to counteract the transmission of virus, which also brings difficulties to monitor large groups of people wearing masks. In this article, we primarily focus on the AI techniques of masked facial detection and related datasets. We survey the recent advances, beginning with the descriptions of masked facial detection datasets. A total of 13 available datasets are described and discussed in detail. Then, the methods are roughly categorized into two classes: conventional methods and neural network-based methods. The conventional methods are usually trained by boosting algorithms with hand-crafted features, which accounts for a small proportion. Neural network-based methods are further classified as three parts according to the number of processing stages. Representative algorithms are described in detail, coupled with some typical techniques that are described briefly. Finally, we summarize the recent benchmarking results, give the discussions on the limitations of datasets and methods, and expand future research directions. To our knowledge, this is the first survey about masked facial detection methods and datasets. Hopefully our survey could provide some help to fight against epidemics.

My Eyes Are Up Here: Promoting Focus on Uncovered Regions in Masked Face Recognition

Abstract: The recent Covid-19 pandemic and the fact that wearing masks in public is now mandatory in several countries, created challenges in the use of face recognition systems (FRS). In this work, we address the challenge of masked face recognition (MFR) and focus on evaluating the verification performance in FRS when verifying masked vs unmasked faces compared to verifying only unmasked faces. We propose a methodology that combines the traditional triplet loss and the mean squared error (MSE) intending to improve the robustness of an MFR system in the masked-unmasked comparison mode. The results obtained by our proposed method show improvements in a detailed step-wise ablation study. The conducted study showed significant performance gains induced by our proposed training paradigm and modified triplet loss on two evaluation databases.

FocusFace: Multi-task Contrastive Learning for Masked Face Recognition

Abstract: SARS-CoV-2 has presented direct and indirect challenges to the scientific community. One of the most prominent indirect challenges advents from the mandatory use of face masks in a large number of countries. Face recognition methods struggle to perform identity verification with similar accuracy on masked and unmasked individuals. It has been shown that the performance of these methods drops considerably in the presence of face masks, especially if the reference image is unmasked. We propose FocusFace, a multi-task architecture that uses contrastive learning to be able to accurately perform masked face recognition. The proposed architecture is designed to be trained from scratch or to work on top of state-of-the-art face recognition methods without sacrificing the capabilities of a existing models in conventional face recognition tasks. We also explore different approaches to design the contrastive learning module. Results are presented in terms of masked-masked (M-M) and unmasked-masked (U-M) face verification performance. For both settings, the results are on par with published methods, but for M-M specifically, the proposed method was able to outperform all the solutions that it was compared to. We further show that when using our method on top of already existing methods the training computational costs decrease significantly while retaining similar performances. The implementation and the trained models are available at GitHub.