Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Neural Best-Buddies in PyTorch

This is our PyTorch implementation for the Neural-Best Buddies paper.

The code was written by Kfir Aberman and supported by Mingyi Shi.

Neural Best-Buddies: Project | Paper

If you use this code for your research, please cite:

Neural Best-Buddies: Sparse Cross-Domain Correspondence Kfir Aberman, Jing Liao, Mingyi Shi, Dani Lischinski, Baoquan Chen, Daniel Cohen-Or, SIGGRAPH 2018.

Prerequisites

  • Linux or macOS
  • Python 2 or 3
  • CPU or NVIDIA GPU + CUDA CuDNN
  • Pytorch > (1.x.x)

Run

  • Run the algorithm (demo example)
#!./script.sh
python3 main.py --datarootA ./images/original_A.png --datarootB ./images/original_B.png --name lion_cat --k_final 10

The option --k_final dictates the final number of returned points. The results will be saved at ../results/. Use --results_dir {directory_path_to_save_result} to specify the results directory.

Output

Sparse correspondence:

  • correspondence_A.txt, correspondence_B.txt
  • correspondence_A_top_k.txt, correspondence_B_top_k.txt

Dense correspondence (densifying based on MLS):

  • BtoA.npy, AtoB.npy

Warped images (aligned to their middle geometry):

  • warp_AtoM.png, warp_BtoM.png

Tips

  • If you are running the algorithm on a bunch of pairs, we recommend to stop it at the second layer to reduce runtime (comes at the expense of accuracy), use the option --fast.
  • If the images are very similar (e.g, two frames extracted from a video), many corresponding points might be found, resulting in long runtime. In this case we suggest to limit the number of corresponding points per level by setting --k_per_level 20 (or any other desired number)

Image Morphing

The morphing results in the paper are based on the paper Automating image morphing using structural similarity on a halfway domain by Liao et al.

Here are detailed instructions for how to combine the output of our code with the image morphing code:

First, please download the exe file from here and test whether it works on your machine with the given demo case as described in the UserGuide.dox or simply run “runme.bat”. Alternatively, the source code can be found here. When it works:

  1. Make a folder e.g. “case1”, containing three files in it: the two input images (should be with the same size) called image1.png and image2.png respectively, and a setting file called “settings.xml”. You can copy the setting.xml form “baby” and modify it to the points that were extracted by our code. For example:

<points image1="0.273438 0.643973 0.657366 0.561384 0.791295 0.641741 0.768973 0.757812 0.309152 0.989955 0.775442 0.994469 0.373051 0.620267 0.420898 0.452148 " image2="0.237832 0.686947 0.627212 0.448009 0.755531 0.625000 0.722345 0.790929 0.308628 0.990044 0.775442 0.994469 0.339644 0.586860 0.420898 0.452148 "/>

where the format is:

<points image1="x1(in image1) y1(in image1) x2(in image1) y2(in image1) x3(in image1) y3(in image1) ……" image2=" x1(in image2) y1(in image2) x2(in image2) y2(in image2) x3(in image2) y3(in image2) "/>

  1. Make a subfolder called “0” under the main folder “case1”. Put two mask images “mask1.png” “mask2.png” that have the same size as the input images and set all the 3 channels to zero.

  2. Run ImageMorphing.exe case1\settings.xml -auto

Citation

If you use this code for your research, please cite our paper:

@article{aberman2018neural,
  title={Neural best-buddies: Sparse cross-domain correspondence},
  author={Aberman, Kfir and Liao, Jing and Shi, Mingyi and Lischinski, Dani and Chen, Baoquan and Cohen-Or, Daniel},
  journal={ACM Transactions on Graphics (TOG)},
  volume={37},
  number={4},
  pages={69},
  year={2018},
  publisher={ACM}
}

About

An algorithm for sparse semantic correspondence between images of different categories [SIGGRAPH 2018]

Resources

License

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •