Daniel Brignac

Daniel Brignac

PhD student at University of Arizona Eletrical and Computer Engineering department studying under Prof. Abhijit Mahalanobis

Publications

Joint Classification and Unknown Detection using Class Conditional Probability Calibration

Published in International Symposium on Signals, Circuits and Systems, 2025

Open-Set Recognition (OSR) addresses the challenge of deployed models encountering unknown classes beyond their fixed training set, often by leveraging known unknowns, though this is not always feasible. We propose combining the decision confidences of a softmax cross-entropy network and a tuplet-loss class-anchor network, achieving superior performance across OSR benchmarks by accurately classifying known samples while reliably rejecting unknowns.

Recommended citation: D Brignac, A Cuellar, B Latibari, A Mahalanobis, "Joint Classification and Unknown Detection using Class Conditional Probability Calibration,"2025 International Symposium on Signals, Circuits and Systems.

CrossModalityDiffusion: Multi-Modal Novel View Synthesis with Unified Intermediate Representation

Published in WACV GeoCV Workshop, 2025

CrossModalityDiffusion is a modular framework designed to address challenges in interpreting geometry across diverse geospatial imaging modalities, such as EO, SAR, and LiDAR. By employing modality-specific encoders and volumetric rendering techniques, it generates geometry-aware feature volumes that unify inputs from varying viewpoints.

Recommended citation: Berian, A., Brignac, D., Wu, J., Daba, N., & Mahalanobis, A. CrossModalityDiffusion: Multi-Modal Novel View Synthesis with Unified Intermediate Representation. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 2025.

Simultaneous Classification of Objects with Unknown Rejection Using Infra-Red Imagery

Published in Sensors, 2025

This method enhances the ability of pretrained classifiers to detect and reject unknown objects in infra-red images without altering their performance on known classes. It introduces a secondary network trained using regression to produce a class-conditional score, indicating whether an input belongs to a known class.

Recommended citation: Cuellar, A.; Brignac, D.; Mahalanobis, A.; Mikhael, W. Simultaneous Classification of Objects with Unknown Rejection (SCOUR) Using Infra-Red Sensor Imagery. Sensors 2025, 25, 492.

Cascading Unknown Detection with Known Classification for Open Set Recognition

Published in International Conference on Image Processing, 2024

Deep learners tend to perform well when trained under the closed set assumption but struggle when deployed under open set conditions. This motivates the field of Open Set Recognition (OSR) in which we seek to give deep learners the ability to recognize whether a data sample belongs to the known classes trained on or comes from the surrounding infinite world. In this work, we decompose the traditional OSR formulation into fine class distinction and known/unknown class discrimination.

Recommended citation: Brignac, Daniel and Mahalanobis, Abhijit, "Cascading Unknown Detection with Known Classification for Open Set Recognition,"2024 International Conference on Image Processing.

Improving Replay Sample Selection and Storage for Less Forgetting in Continual Learning

Published in ICCV Workshops, 2023

Continual Learners commonly employ replay as a method for mitigating catastrophic forgetting where we store few previous examples in a memory buffer and replay them when learning a new task. Replay commonly uses random sampling strategies to populate the buffer which can potentially store uninformative/redundant data and overwrite informative data. This works explore replacements to reservoir sampling for less forgetting when using replay methods.

Recommended citation: Brignac, Daniel and Lobo, Niels and Mahalanobis, Abhijit, "Improving Replay Sample Selection and Storage for Less Forgetting in Continual Learning,"2023 Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops.

Utilization of Data Augmentation Techniques to Enhance Learning with Sparse Datasets

Published in ai4i, 2022

Neural network-based object detection has many important applications but requires a vast amount of training data. In applications where training data may be scarce, data augmentation techniques can be used to expand the training set. This paper explores the performance of such techniques on You Only Look Once Version 5 (YOLOv5).

Recommended citation: R. Yarnell, D. Brignac, Y. Fu and R. F. DeMara, "Utilization of Data Augmentation Techniques to Enhance Learning with Sparse Datasets," 2022 5th International Conference on Artificial Intelligence for Industries (AI4I), Laguna Hills, CA, USA, 2022, doi: 10.1109/AI4I54798.2022.00025.