Fork me on GitHub

Publications

ImageCVPR Pablo Arbelaez*, Jordi Pont-Tuset*, Jonathan T. Barron, Ferran Marques, Jitendra Malik
Multiscale Combinatorial Grouping
Computer Vision and Pattern Recognition (CVPR), 2014
[PDF] [BibTeX] 
@inproceedings{APBMM2014,
  author = {Arbel\'{a}ez, P. and Pont-Tuset, J. and Barron, J. and Marques, F. and Malik, J.},
  title = {Multiscale Combinatorial Grouping},
  booktitle = {Computer Vision and Pattern Recognition},
  year = {2014}
}
ImageArxiv Jordi Pont-Tuset*, Pablo Arbelaez*, Jonathan T. Barron, Ferran Marques, Jitendra Malik
Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation
arXiv:1503.00848, March 2015
[PDF] [arXiv] [BibTeX] 
@inproceedings{PABMM2015,
  author = {Pont-Tuset, J. and Arbel\'{a}ez, P. and Barron, J. and Marques, F. and Malik, J.},
  title = {Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation},
  booktitle = {arXiv:1503.00848},
 month = {March},
  year = {2015}
}

Results

Segmented Object Proposals: Recall at different Jaccard levels

Percentage of annotated objects for which there is a proposal whose overlap with the segmented ground-truth shapes (not boxes) is above J = 0.5, J = 0.7, and J = 0.85, for different number of proposals per image. Results on SegVOC12, SBD, and COCO. Results

Bounding-Box Proposals: Recall at different Jaccard levels

Percentage of annotated objects for which there is a bounding box proposal whose overlap with the ground-truth boxes is above J = 0.5, J = 0.7, and J = 0.85, for different number of proposals per image. Results on SegVOC12, SBD, and COCO. Results

Qualitative results on COCO images

Image, ground truth, multi-scale UCM and best MCG proposals among the 500 best ranked. Qualitative

Code


The code contains the four following versions or packages:

MCG benchmark
Tools to benchmark algorithms that generate segmented object candidates. We propose two different measures (jaccard index at instance and class levels) which we sweep against the number of proposed candidates. The code is in Matlab with some parts in C++ pre-compiled for Linux, Windows, and Mac.

MCG pre-trained
Code to compute MCG candidates and hierarchies (UCMs) with models pre-trained on the BSDS500 and the PASCAL 2012 segmentation datasets (im2mcg and im2ucm functions). It comes pre-compiled for Linux and Mac and it is not compatible with Windows.

MCG full
Full code to re-train MCG (Pareto training, random forest ranking, etc.) on new datasets and on different object categories. The hierarchies at multiple scales should be re-computed before training on new datasets. It comes pre-compiled for Linux and Mac and it is not compatible with Windows.

DNCuts
Stand-alone Matlab code for fast eigenvector computation in Normalized Cuts segmentation.

Datasets

MCG (Multiscale Combinatorial Grouping) SCG (Single-scale Combinatorial Grouping)
Proposals UCMs Boxes Proposals UCMs Boxes
BSDS500
train+val+test: 500 im.
Pascal Segm. 2012
train+val+test: 4369 im.
Pascal Main 2012
train+val+test: 22531 im.
Pascal Main 2007
train+val+test: 9963 im.
SBD
train+val: 12031 im.
MS COCO
train 2014: 82783 im.
MS COCO
val 2014: 40504 im.
MS COCO
test 2014: 40775 im.
MS COCO
test 2015: 81434 im.
ILSVRC Det. 2013
val: 20121 im.
ILSVRC Det. 2013
test: 40152 im.