During my Ph.D. period, I worked on 3D reconstruction of multi-source data including satellite, aerial, UAV, ground-view images; 3D registration of cross-view/cross-source data; View synthesis & generation of ground views from aerial & satellite views.
SkyEyes is a framework that transforms aerial imagery into realistic street view sequences using 3D Gaussian Splatting, diffusion models, and a constrained optimization strategy to enhance cross-view synthesis quality.
The first work applying diffusion-based method to tackle satellite-to-ground view generation task. It performs ground-view synthesis conditioning on the weak building facades information from satellite images.
Want ICP (iterative closes point) be applied to terrain-scale DSMs (digital surface model) and even multiple noisy DSMs? Check out the proposed DSM-ICP, which applies a fast and exact nearest neighbor search method leveraging the grid structure of DSM.
NeRF vs Multi-view stereo? We propose multi-camera tiling technique to enable NeRF on large-scale aerial datasets and further conduct experiment to compare their geometry reconstruction performance.
MCT-NeRF is selected as the Cover article of 12/2024 issue, The Photogrammetry Record
Academic Activity
Invitied Talks
Closing the Gap Between Satellite and Street-View Imagery Using Generative Models Voxel51 ECCV 2024 Redux, Nov 21, 2024
Fraunhofer Heinrich Hertz Institute, German Cancer Research Center, Heidelberg University
Reviewer:
Neural Information Processing Systems (NeurPIS) 2024
British Machine Vision Conference (BMVC) 2024
Asian Conference on Computer Vision (ACCV) 2024
ISPRS Journal of Photogrammetry and Remote Sensing
IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)