Complementary vision · Generative reconstruction
CoGenRec: A Unified Generative Framework for Complementary Vision Reconstruction
1 Department of Precision Instrument, Tsinghua University
2 Pen-Tung Sah Institute of Micro-Nano Science and Technology, Xiamen University
3 The Hong Kong University of Science and Technology
4 Primevision Technology
Overview
Reconstructing sharp, high-frame-rate HDR video from complementary RGB and spatio-temporal measurements.
The complementary vision sensor captures spatially dense RGB frames alongside high-speed spatial and temporal difference signals. CoGenRec combines these measurements with a video generative prior to jointly address motion blur, temporal undersampling, and exposure clipping.
01
Restore sharp structures
Use intra-exposure cues to recover motion-blurred RGB frames.
02
Recover motion between frames
Use inter-exposure measurements for high-frame-rate reconstruction.
03
Fuse complementary exposure information
Combine the two sensing pathways for HDR reconstruction.
Real-world results.
Ground fireworks
Balloon burst
Train encounter (600 km/h relative speed)
Peach blossoms
Street bicycles
Method comparisons
Spinning disc
Rotating blocks
Cartoon poster
Sunlit facade
Passing motorcycle
Resources
Paper
Read on arXivCode & models
Citation
@misc{meng2026cogenrecunifiedgenerativeframework,
title={CoGenRec: A Unified Generative Framework for Complementary Vision Reconstruction},
author={Yapeng Meng and Yuguo Chen and Xiangru Chen and Shaojun Xu and Xuezhen Wang and Lijian Wang and Zheyu Yang and Yihan Lin and Luping Shi and Rong Zhao},
year={2026},
eprint={2611.07665},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2611.07665},
}