Hello There!
- Full Name:Rosaura G. VidalMata
- Pronunciation: [ro-SAW-rah] [bee-DAHL]
- Email:rosauravidal@gmail.com
- Website:www.rvidal.me
- Research Interests:Edge Model Optimization, Image Enhancement, Real-Time Perception, and Computer Vision.
I specialize in edge model optimization, real-time neural perception, and visual recognition under severe constraints. My work focuses on deploying efficient deep learning models across mobile and edge platforms (using PyTorch, C++, and TFLite) spanning low-light image enhancement, stereo depth, and document perception. Actively engaged in the research community, I regularly publish first-author work in top venues including IEEE TPAMI and WACV, and co-organize benchmark challenges advancing computer vision in adverse and unconstrained environments.
My Resume
-
Education
-
Ph.D. in Computer Science and Engineering
University of Notre Dame - 2024Thesis topic: "Rich Embedding Techniques to Improve Scene Understanding"
Advisors:Kevin W. Bowyer, Walter J. Scheirer. GPA: 3.95
-
M.S. in Computer Science and Engineering
University of Notre Dame - 2020GPA: 3.95
-
Bachelor of Science in Computer Science and Engineering
Tecnologico de Monterrey - 2015Excellence Scholarship. GPA: 4.0
-
Work Experience
-
AI Researcher
Lenovo Research - April 2022 - Present- Led cross-functional technical delivery of a dual-camera video pipeline showcased at NVIDIA GTC 2026, adapting stereo foundation models for hybrid edge-cloud execution to accelerate throughput from 0.89 FPS to 27 FPS (30× faster).
- Engineered an on-device document perception framework in C++, curating 52,216 multi-source images to increase AP75 by 30% over baseline and unifying camera and microphone modules into a shared low-latency runtime.
- Optimized a mobile low-light model for Motorola Edge via TFLite using FP16 quantization, layer pruning, and filter tuning, shrinking footprint from 1,554 MB to 1,198 MB (1.3× smaller) and runtime from 4.42 s to 1.89 s (2.3× faster) while meeting DXOMARK exposure and color benchmarks.
- Built an automated annotation and curation pipeline that raised throughput 40% and production output 130%, compiling and releasing the LENVIZ low-light benchmark (230,000+ frames; WACV 2026).
-
Graduate Research Intern
MIT-IBM Watson AI Lab - June - August 2019- Developed a joint visual-temporal embedding for unsupervised action segmentation in untrimmed video, gaining 4% mean-over-frames (MoF) over published state of the art on the Breakfast Actions dataset (published at WACV 2021).
- Won Best of Show at the IBM intern showcase, out of 50+ projects.
-
Doctoral Researcher, Computer Science & Engineering
University of Notre Dame - June 2016 - January 2024- Released UG2, a benchmark of 150,000+ aerial and ground video frames under weather, distance, and blur, reported in TPAMI and WACV (funded by IARPA and NVIDIA).
- Co-organized the CVPR 2018 and 2019 UG2+ prize challenges, evaluating 40 international algorithm submissions on video object recognition under sensor and weather degradation to publish foundational benchmarks in IEEE TPAMI and CVIU.
- Benchmarked 10 manipulation-localization methods and released an open-source detection toolkit under DARPA funding, designing a residual preprocessor with anomaly-enhancement loss and disparity augmentation for social-media compression.
-
Honors & Awards
-
Women in Computer Vision Feature
RSIP Vision @ WACV - 2026Selected and featured in the "Women in Computer Vision" series in the WACV 2026 Daily magazine by RSIP Vision.
-
Lenovo Research SVP Award of Individual Excellence
Lenovo Research - 2023Honored for driving breakthrough computer vision research and key deliverable milestones.
-
Best of Show
IBM Intern Showcase - 2019Won "Best of Show" across 50+ intern projects at the MIT-IBM Watson AI Lab for visual-temporal action segmentation.
Publications
M. Aithal, R. G. VidalMata, M. Kartha, G. Chen, E. Adhikarla, L. N. Kirsten, Z. Fu, N. A. Madhusudhana, J. Nasti, "LENVIZ: A High-Resolution Low-Exposure Night Vision Benchmark Dataset," in 2026 IEEE Winter Conference on Applications of Computer Vision (WACV).
S. Banerjee*, R. G. VidalMata*, Z. Wang, and W. J. Scheirer, "Report on UG2+ Challenge Track 1: Assessing Algorithms to Improve Video Object Detection and Classification from Unconstrained Mobility Platforms," in Computer Vision and Image Understanding (CVIU).
S. Abraham, Z. Carmichael, S. Banerjee, R. VidalMata, A. Agrawal, M. N. A. Islam, W. Scheirer, and J. ClelandHuang, "Adaptive Autonomy in Human-on-the-Loop Vision-Based Robotics Systems," in 2021 Workshop on AI Engineering (WAIN).
R. G. VidalMata, W. J. Scheirer, A. Kukleva, D. D. Cox, and H. Kuehne, "Joint Visual-Temporal Embedding for Unsupervised Learning of Actions in Untrimmed Sequences," in 2021 IEEE Winter Conference on Applications of Computer Vision (WACV)
R. G. VidalMata, [et. al.], "Bridging the Gap Between Computational Photography and Visual Recognition," in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), doi: 10.1109/TPAMI.2020.2996538.
R. G. VidalMata, S. Banerjee, W. J. Scheirer, K. Grm, and V. Struc, "UG2: a Video Benchmark for Assessing the Impact of Image Restoration and Enhancement on Automatic Visual Recognition," in 2018 IEEE Winter Conference on Applications of Computer Vision (WACV), pp. 1597-1606, March 2018. doi 10.1109.
A. Czajka, K. W. Bowyer, M. Krumdick, and R. G. VidalMata, "Recognition of Image-Orientation-Based Iris Spoofing," in IEEE Transactions on Information Forensics and Security, vol. 12, no. 9, pp. 2184-2196, Sept. 2017. doi: 10.1109/TIFS.2017.2701332
Research Projects
Low-Light Image Enhancement & The LENVIZ Benchmark
Capturing high-fidelity imagery in extreme low-light environments is a fundamental challenge for edge platforms due to sensor noise, limited dynamic range, and motion artifacts.
This initiative developed on-device restoration and alignment models under strict performance budgets, alongside curating the LENVIZ benchmark: a 230,000-frame dataset (WACV 2026) advancing nighttime visual recognition.
Disparity Augmentation for Manipulation Detection
Manipulation detection algorithms often rely on identifying local anomalies, where manipulated regions would be “sufficiently” different from the rest of the features in the image.
As part of this project we study techniques taken from computational photography as a way to exacerbate the anomalies present in manipulated regions to facilitate their detection by a variety of deep-learning and traditional manipulation detection methods.
Human-Drone Partnerships for Emergency Response Scenarios
The use of small Unmanned Aerial Vehicles to collect imagery in difficult or dangerous terrain offers clear advantages for time-critical tasks such as search-and-rescue missions, fire surveillance, and medical deliveries.
Employing a drone to search for a missing kayaker on a river or a child lost in the wilderness, survey a traffic accident or a forest fire, or to track a suspect in a school shooting would not only reduce risk to first responders but also allow for a wide-scale search to be deployed in an expedited manner.
Unsupervised Temporal Segmentation of Actions
Understanding the structure of complex activities in untrimmed videos is a challenging task in the area of action recognition. One problem here is that this task usually requires a large amount of hand-annotated minute- or even hour-long video data, but annotating such data is very time consuming and can not easily be automated or scaled.
To address this problem, we propose an approach that is able to provide a meaningful visual and temporal embedding out of the visual cues present in contiguous video frames.
Image Restoration and Enhancement for Visual Recognition
Can the application of enhancement algorithms as a pre-processing step improve image interpretability for manual analysis or automatic visual recognition to classify scene content?
While there have been important advances in the area of computational photography to restore or enhance the visual quality of an image, the capabilities of such techniques have not always translated in a useful way to visual recognition tasks.
Iris Image Orientation Recognition
Matching the iris codes from the left and right eyes of the same person gives a result that is on average basically the same as matching iris codes from unrelated persons.
Two approaches are compared on the same data, using the same evaluation protocol:
- Feature engineering, using hand-crafted features classified by a support vector machine (SVM)
- Feature learning, using data-driven features learned and classified by a convolutional neural network (CNN)