Aymen Sekhri
PhD researcher in machine learning and computer vision.
Cotutelle between XLIM, University of Poitiers and Colourlab, NTNU.
I build models that measure what people actually see. My work sits between perceptual science and efficient deep learning: image quality metrics for augmented reality, and trustworthy computer vision for medical imaging.
Augmented reality breaks most of the assumptions image quality metrics are built on. The viewport is a composite — a captured real scene underneath, a rendered virtual layer on top — and the two interfere perceptually in a way that no reference image can describe. I work on blind metrics that handle this directly, and on making them small enough to run on the device doing the rendering.
The same question shows up in medical imaging from the other side: when a model’s output feeds a diagnosis, you need to know which local structures it is actually looking at.
Research directions
- Blind quality assessment for AR — metrics for composited real/virtual content, including the distortions and visual confusion unique to immersive pipelines.
- Perceptual supervision — training on human ranking feedback rather than absolute scores, because that is what observers are consistent at.
- Efficient perceptual models — knowledge distillation and compact transformers that keep perceptual accuracy at a fraction of the compute.
- Trustworthy medical imaging AI — localisation-aware architectures and domain adaptation for radiographic severity grading.
Selected publications
All 13 →-
When Restoration Becomes the Reference: Reusing Full-Reference IQA in Blind Settings
conference IEEE International Conference on Image Processing (ICIP)
-
ARIQA-3DS: A Stereoscopic Image Quality Assessment Dataset for Realistic Augmented Reality
dataset arXiv preprint · 5th Workshop on Computer Vision for Multimedia Spatial Intelligence
-
Enhancing Content Representation for AR Image Quality Assessment Using Knowledge Distillation
journal IEEE Transactions on Circuits and Systems for Video Technology
-
TransformAR: A Light-Weight Transformer-Based Metric for Augmented Reality Quality Assessment
journal Signal Processing: Image Communication, vol. 140, art. 117437
-
ARaBIQA: A Novel Blind Image Quality Assessment Model for Augmented Reality
conference IEEE International Conference on Image Processing (ICIP), pp. 379–384
News
-
ICIP 2026 — presenting When Restoration Becomes the Reference, on reusing full-reference IQA metrics in blind settings.
-
ARIQA-3DS is out — the first large stereoscopic AR image quality dataset, 1,200 viewports rated by 36 participants in a video see-through HMD.
-
Our knowledge-distillation work on AR image quality is accepted at IEEE TCSVT, and TransformAR appears in Signal Processing: Image Communication.
-
ARaBIQA presented at ICIP 2025 — a blind quality model built around AR-specific distortions.
-
Ranking-guided lightweight quality prediction presented at ICASSP 2025.
Get in touch
I am co-advised by Prof. Mohamed-Chaker Larabi (University of Poitiers) and Prof. Seyed Ali Amirshahi (NTNU). I am happy to hear from anyone working on AR/VR quality, perceptual modelling, or applied machine learning for medical imaging — collaborations, reviews, or just a question about one of the papers.