SplineOps: A High-Performance Spline Library for Signal and Image Processing
Pablo Garcia-Amorena, Scientific Assistant at EPFL

Meeting • 2025-11-25

More Info ...Abstract
SplineOps (https://splineops.github.io/) is an open-source Python and C++ library that turns spline theory into a high-performance engine for signal and image processing. Built on decades of work at EPFL’s Biomedical Imaging Group, it provides efficient B-spline models for interpolation, multidimensional resizing, geometric transforms, smoothing, regression, multiscale analysis, and differential operators. Thanks to a new highly optimized resizing core, SplineOps outperforms widely used state-of-the-art libraries (SciPy, PyTorch, OpenCV, Pillow, scikit-image) on realistic resize round-trip benchmarks, simultaneously improving image quality and speed. I will present the main modules and benchmark results, explain the design choices that make this performance jump possible, and validate performance across broader datasets, applications, and computing platforms.