Workshops Archive

Utilizing Deep Learning-Based Video Frame Interpolation for Scientific Visualization


Workshop: ISAV26: In Situ AI, Analysis and Visualization

Authors: Jonas Lukasczyk and Christoph Garth (RPTU University) and Ethan Vasey Stam, David Honegger Rogers, and Divya Banesh (Los Alamos National Laboratory)

Abstract: Scientific visualization workflows must balance visual fidelity against the storage and I/O costs of large image databases. We investigate deep learning-based Video Frame Interpolation (VFI) to reduce these costs by reconstructing omitted visualization images from stored keyframes. We evaluate the FILM and RIFE models on scientific visualization images using perceptual quality metrics and integrate them into the PyCinema analysis framework. Based on these results, we develop a greedy keyframe selection algorithm that stores an image only when it cannot be reconstructed from neighboring keyframes within a user-defined error tolerance. Its reliance on local information enables integration into an in situ ParaView Catalyst workflow. Experiments show that VFI can accurately reconstruct omitted images while substantially reducing the number of frames written to disk, improving the storage efficiency of scientific visualization pipelines.


Back to ISAV26: In Situ AI, Analysis and Visualization Archive Listing

Back to Full Workshop Archive Listing