ISET-LFM: A Physics-based Synthetic Radiance Dataset for LED Flicker Mitigation in Automotive Imaging

This dataset provides high-fidelity, camera-agnostic scene radiance sequences designed to support the development and benchmarking of LED Flicker Mitigation (LFM) algorithms. While modern High Dynamic Range (HDR) automotive sensors utilize multi-exposure captures to manage high-contrast scenes, the interaction between pulsed LED sources and short exposure duration often results in missing signals (flicker). Conversely, long exposure duration captures the signal but introduces significant motion-induced blur.
To address this motion-flicker trade-off, this dataset contains 200 diverse synthetic automotive scenes hand-picked for the prominence of LED sources. Each scene provides simultaneous short-exposure (3–5 ms) and long-exposure (11.11 ms) spectral radiance maps rendered with non-uniform motion blur corresponding to ego-velocities between 20–50 m/s.
By providing decomposed light groups (headlights, streetlights, environment, and others), the dataset enables researchers to modulate LED parameters (duty cycle and frequency) of individual light groups without re-rendering. This resource also serves as a foundation for training data-driven models to reconstruct sharp, flicker-free images for safety-critical Advanced Driver Assistance Systems (ADAS).
More details about organization of the dataset can be found in README.md.
This repository contains data files that were created by code and techniques detailed at https://github.com/AyushJam/iset-lfm