Skip to content

A Matlab implementation for the paper: "Learning 3D Deformation of Animals from 2D Images" http://www.umiacs.umd.edu/~kanazawa/papers/cat_eg2016.pdf

Notifications You must be signed in to change notification settings

akanazawa/catdeform

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 

Repository files navigation

Learning 3D Deformation of Animals from 2D Images

A Matlab implementation of the paper "Learning 3D Deformation of Animals from 2D Image"

Steps

  1. Get this repo: git clone [email protected]:akanazawa/catdeform.git
  2. Download data from this link, it contains the 10 images of horses and cats along with their annotations.
  3. Download toolbox from this link
  4. Untar them, put data/ and toolbox/ in the project root so it is in the same directory as this README.md
  5. Make sure you have the dependencies as specified below & change the path in code/initialize.m accordingly
  6. Start matlab in code/
  7. Try:
    • script_joint_cat/script_joint_horse for solving for the deformation & stiffness over 10 cats/horses (as in the paper)
    • script_cat : Initial settings is on Uniform (no stiffness learning) solving for one cat image (best for testing your environment is setup properly), can be modified for reusing learned stiffness (see the file)
    • script_run_many_cat/script_run_many_horse for the ablation study

Results in the paper

Can be downloaded from this link. You need to untar and put results/ directory in the project root (along with this README), or compute it yourself for re-use stiffness experiments (that takes in pre-computed stiffness & solves for deformation).

The results/ include results for 10 cats and horses with stiffness shown in the paper. For cats there the results obtained with UNIFORM bound setting is also supplied (those under prefix uniformC). Those for horse can be computed by running script_run_many_horse.

Dependencies:

Requires MATLAB, tested on 64-bit Linux: Ubuntu 14.04, MATLAB R2015a.

Install MOSEK & change the path in initialize.m accordingly. Make sure to use the latest version of YALMIP (included in toolbox 2015/09/19 release) otherwise the optimization will be slower.

Windows users: Untested but you can probably use it if pre-compiled binaries of iso2mesh in toolboxes/iso2mesh/bin/ and sba matlab (in toolboxes/sba_1.5/matlab) work.

OS X users: Binaries of iso2mesh didn't work for me on El Capitan and couldn't compile SBA on OS X, but if you really want to make it work please compile them by following the information on iso2mesh and sba.

Compiling SBA

In case sba.mexa64 doesn't work fro you, compile it by:

You need to got to toolbox/sba_1.5/matlab

and do

make

It requires BLAS and LAPACK. What worked for me requires these packages:

sudo apt-get install libblas-dev f2c liblapack-pic

Change the paths to your libraries in toolbox/sba_1.5/matlab/Makefile accordingly.

Notes:

  • code for rendering results is in code/util/figure_code
  • Use vtk to visualize the result saved from each iteration, or convert it to .off using code/util/vtk2off

Bibtex

Please cite our papaer if this code was useful to you.

@inproceedings{Kanazawa:2016,
author = {Angjoo Kanazawa and Shahar Kovalsky and Ronen Basri and David W. Jacobs},
title = {Learning 3D Deformation of Animals from 2D Images},
booktitle = {Proceedings of EUROGRAPHICS/ACM SIGGRAPH Symposium on Geometry Processing},
year = {2016},
}

Disclamer:

The code is provided as-is for academic use only and without any guarantees. Please contact the author to report any bugs. Written by Angjoo Kanazawa based on Controlling Singular Values code by Shahar Kovalsky and Noam Aigerman

About

A Matlab implementation for the paper: "Learning 3D Deformation of Animals from 2D Images" http://www.umiacs.umd.edu/~kanazawa/papers/cat_eg2016.pdf

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages