"GrabCut": interactive foreground extraction using iterated graph cuts

By Carsten Rother, Vladimir Kolmogorov, Andrew Blake. 2004

6,743 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

COCO-Stuff: Thing and Stuff Classes in Context (2016, methodology): “Several works use interactive segmentation methods [42, 57, 10] to speedup annotation; others annotate superpixels [61, 22, 40].”
DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (2016, methodology): “CRFs have been broadly used in semantic segmentation to combine class scores computed by multi-way classifiers with the low-level information captured by the local interactions of pixels and edges [23], [24] or superpixels [25].”
SimpleClick: Interactive Image Segmentation with Simple Vision Transformers (2022, methodology): “We report results on ﬁve benchmarks: GrabCut [40], Berkeley [37], SBD [20], DAVIS [39], and Pascal VOC [14].”
FocalClick: Towards Practical Interactive Image Segmentation (2022, methodology): “Besides, GrabCut and Berkeley only contain 50 and 100 easy examples, making the results not convincing.”
Reviving Iterative Training with Mask Guidance for Interactive Segmentation (2021, methodology): “Early methods [6] tackled the problem with optimization-based approaches minimizing a cost function defined on a graph over image pixels.”
Iteratively Trained Interactive Segmentation (2018, methodology): “To reach the high threshold of 90% on GrabCut [25], our method needs slightly more clicks than RIS-Net [17].”
Deep Extreme Cut: From Extreme Points to Object Segmentation (2017, methodology): “Our method is extensively validated in ﬁve publicly available databases: PASCAL [8], COCO [21], DAVIS-2016 [27], DAVIS-2017 [31], and Grabcut [32], for various experimental setups that show its applicability and generalization capabilities.”
Extreme Clicking for Efficient Object Annotation (2017, methodology): “Similar to [51], we employ a binary pairwise energy function E defined over the pixels and their labels.”
Deep Interactive Object Selection (2016, methodology): “We evaluate all the algorithms on four public datasets: Pascal VOC 2012 segmentation validation set, Grabcut [19], Berkeley [16] and MS COCO [13].”
Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs (2014, methodology): “Traditionally, conditional random ﬁelds (CRFs) have been employed to smooth noisy segmentation maps (Rother et al., 2004; Kohli et al., 2009).”
BoxSup: Exploiting Bounding Boxes to Supervise Convolutional Networks for Semantic Segmentation (2015, methodology): “Further, the adopted low-level features (like color statistics [23]) may limit the quality of the segmentation results.”
Weakly-and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation (2015, methodology): “This approach is similar to the grabcut method of [33].”
Higher Order Conditional Random Fields in Deep Neural Networks (2015, methodology): “GrabCut [42]), and represents a crude segmentation of the detected object.”
Articulated Pose Estimation by a Graphical Model with Image Dependent Pairwise Relations (2014, methodology): “Conditional Random Fields (CRFs): Our model is also related to the conditional random field literature on data-dependent priors [18, 13, 15, 19].”
What's the Point: Semantic Segmentation with Point Supervision (2015, background): “tion cost through effective interfaces [30, 21, 22, 41, 3], e.”
Iterative Instance Segmentation (2015, methodology): “More sophisticated approaches model the problem using CRFs with unary and pairwise potentials [27, 24, 19].”
Segmentation from Natural Language Expressions (2016, methodology): “Also, it can be seen that using GrabCut [9] to segment the foreground from bounding boxes (“SCRC grabcut” and “GroundeR grabcut”) results in higher precision for both SCRC and GroundeR than using the entire bounding box region.”
Instance-Level Segmentation for Autonomous Driving with Deep Densely Connected MRFs (2015, methodology): “In interactive segmentation, approaches like Grabcut [24, 4] require a user-supplied box, or a scribble on the foreground and background, in order to segment the objects.”
Bottom-up Instance Segmentation using Deep Higher-Order CRFs (2016, methodology): “} The foreground is obtained from a foreground/background segmentation method such as GrabCut [30] and indicates a rough segmentation of the detected object.”