Supporting Dataset for VSRNet — Part 1: Reference Frames, Vibration-Degraded Inputs, and Evaluation Materials

Supporting data for the manuscript “VSRNet: Lightweight Video Restoration with Temporal Stability and Perception-Oriented Evaluation for Vibration-Degraded Shipboard Vision.” This record is Part 1 of the supporting dataset and contains clean reference frames (GT), synthetically generated vibration-degraded inputs under low, mid, and high vibration levels, and evaluation materials supporting the reconstruction-quality, motion-stability, downstream-perception, and computational-efficiency analyses reported in the manuscript. The final quantitative evaluation uses the common valid frame interval from frame 000002 to frame 002246. These evaluation data are disjoint from the VSRNet training and validation data. Companion Zenodo records contain the restored outputs produced by FastDVDnet, VRT, and VSRNet. Source code is available at https://github.com/920961270/VSRNet.