This is a better and simpler implementation for "SRHEN: Stepwise-Refining Homography Estimation Network via Parsing Geometric Correspondences in Deep Latent Space".
MACE=1.18 on Synthetic COCO dataset. (MACE=9.19 in the original paper, without using the coarse-to-fine framework).
If you find this work useful, please consider citing:
@inproceedings{li2020srhen,
title={SRHEN: Stepwise-Refining Homography Estimation Network via Parsing Geometric Correspondences in Deep Latent Space},
author={Li, Yi and Pei, Wenjie and He, Zhenyu},
booktitle={Proceedings of the 28th ACM International Conference on Multimedia},
pages={3063--3071},
year={2020}
}
"preprocess_images_offline.py"
, i.e., DIR_IMG
and DIR_OUT
according to your own directory."python preprocess_images_offline.py"
."train.py"
, i.e., DIR_IMG
and DIR_MOD
for train images and trained models, respectively."python train.py"
."test.py"
, i.e., DIR_IMG
and DIR_MOD
for test images and saved models, respectively."python test.py"
.Link: https://pan.baidu.com/s/1u-9Bu72OY_wC7CoocorZXQ?pwd=94rq Extraction Code: 94rq