mmaire/ae-multigrid

Multigrid Angular Embedding Eigensolver

This software implements the multigrid eigensolver described in:

[Progressive Multigrid Eigensolvers for Multiscale Spectral Segmentation] (http://ttic.uchicago.edu/~mmaire/papers/pdf/seg_multigrid_iccv2013.pdf)
Michael Maire and Stella X. Yu
International Conference on Computer Vision (ICCV), 2013

It also serves as a core component of the simultaneous segmentation and figure/ground organization system described in:

[Affinity CNN: Learning Pixel-Centric Pairwise Relations for Figure/Ground Embedding] (http://arxiv.org/abs/1512.02767)
Michael Maire, Takuya Narihira, and Stella X. Yu
Computer Vision and Pattern Recognition (CVPR), 2016

Demo - Multiscale Segmentation (ICCV 2013)

Run demo.m from MATLAB for a multiscale image segmentation demo.

Demo - Figure/Ground (CVPR 2016)

Run ae_cnn_demo.m from MATLAB for a segmentation+figure/ground demo.
Note that this demo runs in single-grid compute mode.

See util/bench_fg.m for figure/ground benchmark evaluation code.

Notes

The current implementation is efficient for image sizes (length and width) divisible by 2^(s-1), where s is the number of pyramid levels. See the included multiscale_resize.m function for padding arbitrary images to the nearest efficient size.

ISPC

The ispc/ subdirectory contains a sparse matrix times dense matrix multiply routine that is significantly faster than Matlab's built-in operation on machines supporting the AVX instruction set (most processors released in 2011 or later).

Uncomment use_ispc = 1; in demo.m to use this implementation. It relies on an included precompiled Linux mex file. To compile for other architectures, download ISPC from http://ispc.github.io/ and run build.sh.

Citation

If you make use of this software, please cite the following in any publications:

@inproceedings{MY:ICCV:2013,
    title     = {Progressive Multigrid Eigensolvers for Multiscale Spectral Segmentation},
    author    = {Michael Maire and Stella X. Yu},
    booktitle = {International Conference on Computer Vision (ICCV)},
    year      = {2013}
}

If you make use of our figure/ground component, please also cite:

@inproceedings{MNY:CVPR:2016,
    title     = {Affinity CNN: Learning Pixel-Centric Pairwise Relations for Figure/Ground Embedding},
    author    = {Michael Maire and Takuya Narihira and Stella X. Yu},
    booktitle = {Computer Vision and Pattern Recognition (CVPR)},
    year      = {2016}
}

License

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License along with this program. If not, see http://www.gnu.org/licenses/.

External Dependencies

The image segmentation demo relies on local contour cues computed using the Berkeley contour detector (Pb). A version of this software is included in the grouping/ subdirectory. For more information see:

http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html

mmaire / ae-multigrid