"Vflare"

"VFlare is a dedicated dataset for video lens flare synthesis and removal, designed to address the motion independence between flare artifacts and scene content. Unlike existing image-focused flare datasets, VFlare models dynamic flare behavior over time, including both scattering and reflective components, under physically informed constraints related to light-source motion and optical geometry. The dataset contains large-scale paired synthetic video sequences for supervised training and evaluation, together with real-world flare videos for testing generalization in practical scenarios. By providing temporally coherent degradations and diverse flare patterns, VFlare supports research on spatio-temporal restoration, flare-scene disentanglement, and motion-consistent enhancement. It serves as a benchmark for developing robust video flare removal methods with improved visual quality and temporal stability."