Skip to content

Latest commit

 

History

History
33 lines (27 loc) · 2.17 KB

readme.md

File metadata and controls

33 lines (27 loc) · 2.17 KB
Raw

Zoom and Learn (ZOLE)

This repo includes the test model of the paper: "Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains" by Jiahao Pang, Wenxiu Sun, Chengxi Yang, Jimmy Ren, Ruichao Xiao, Jin Zeng and Liang Lin.

Please cite our paper if you find this repo useful for your work:

@inproceedings{pang2018zoom,
    title = {Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains},
    author = {Pang, Jiahao and Sun, Wenxiu and Yang, Chengxi and Ren, Jimmy and Xiao, Ruichao and Zeng, Jin and Lin, Liang},
    booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2018}
}

Prerequisites

  • Modified Caffe provided by the Computer Vision Group, University of Freiburg link
  • MATLAB (Our scripts has been tested on MATLAB R2015a)
  • Download our trained model through this MEGA link or this Baiduyun link

Testing on Real Stereo Pairs

  • Compile the modified Caffe and its MATLAB interface (matcaffe).
  • Put this repo (with name "zole", for example) in the "dispflownet-release/models" folder.
  • Put the downloaded model, "zole.caffemodel", in the "dispflownet-release/models/zole" folder.
  • Run the MATLAB script "test_zole.m" to test our trained model. We provide 4 real stereo pairs in this repo for testing.

Training

We do not provide our training code due to company regulations. Moreover, the model provided in this repo is different from the one for our in-house usage. In other words, it is a model only for demonstration purpose. If you have interests in this work, you may check out the website of SenseTime.

Results

This is what you should see if you have successfully run our demo code, which shows the left image and the corresponding disparity map.

N|Solid