Renata Khasanova

École Polytechnique Fédérale de Lausanne

Papers

Total Citations

H-Index

About

Renata Khasanova is a researcher advancing computer vision for non-planar imagery, with a focus on omnidirectional image analysis. Her work addresses a critical challenge: conventional convolutional methods, designed for planar images, perform suboptimally on the wide-field-of-view images captured by cameras used in robotics, autonomous vehicles, and virtual reality. Khasanova’s key contribution lies in developing geometry-aware approaches that respect the spherical or cylindrical structure of omnidirectional data. Her most cited paper, “Graph-Based Classification of Omnidirectional Images” (2017, 69 citations), introduces a novel graph-based framework that models pixel relationships on non-Euclidean domains, enabling more accurate classification. She extended this in “Geometry Aware Convolutional Filters for Omnidirectional Images Representation” (2019, 8 citations), designing filters that adapt to the image’s intrinsic geometry, improving representation learning for navigation and scene understanding. By bridging graph theory and geometric deep learning, Khasanova’s work has influenced how autonomous systems process panoramic visual input, offering a principled alternative to naive planar projections. Her research is pivotal for students and engineers working on perception for drones, robots, and VR, demonstrating that accounting for sensor geometry can unlock significant performance gains in real-world computer vision tasks.