MPI-Flow: Learning Realistic Optical Flow with Multiplane Images
Embedded Files
Paper accepted by ICCV 2023
Liang, Yingping and Liu, Jiaming and Zhang, Debing and Ying, Fu
MPI-Flow:
We are the first to present a novel MPI-based optical flow dataset generation framework, namely MPI-Flow, which can significantly improve the realism of the gen erated images and motion.
We present a novel independent object motion module for modeling dynamic objects in MPI, which can model realistic optical flow from camera motion and object motion simultaneously.
We design a depth-aware inpainting module for realistic image inpainting, which can remove unnatural motion occlusions in generated images.
Use MPI-Flow to create continuous images and optical flows from single 2D image!
@inproceedings{liang2023mpi,
author = {Liang, Yingping and Liu, Jiaming and Zhang, Debing and Ying, Fu},
title = {MPI-Flow: Learning Realistic Optical Flow with Multiplane Images},
booktitle = {In the IEEE International Conference on Computer Vision (ICCV)},
year={2023}
}
The camera-ready paper is available at:
Page updated
Google Sites
Report abuse