DenseCut: Densely Connected CRFs for Realtime GrabCut

By Ming-Ming Cheng, Victor Adrian Prisacariu, Shuai Zheng, Philip H. S. Torr, Carsten Rother. 2015

91 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Artificial Intelligence, 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 Aerospace Engineering

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Reviving Iterative Training with Mask Guidance for Interactive Segmentation (2021, background): “For example, one of the simplest types of interactions is a bounding box, that is used in [23, 18, 24].”
Interactive Image Segmentation with Latent Diversity (2018, methodology): “Until recently, prior knowledge about object appearance was encoded primarily in hand-crafted features and distance metrics [4, 37, 12, 2, 9, 13, 35, 5, 24, 8].”
Efficient Likelihood Learning of a Generic CNN-CRF Model for Semantic Segmentation (2015, methodology): “t would be interesting to combine the dense CRF model of [11] with our generic CRF model. Note that a dense CRF is implicitly modeling the property that objects have a compact color distribution, see [39], which is a complementary modeling power to our generic CRF model. Acknowledgements. This work ”
Joint Training of Generic CNN-CRF Models with Stochastic Optimization (2015, background): “Note that a dense CRF is implicitly modeling the property that objects have a compact color distribution, see [39], which is a complementary modeling power to our generic CRF model.”
Weakly Supervised Object Boundaries (2015, methodology): “We also experimented with DenseCut [8] and CNN+GraphCut [29], but did not obtain any gain; thus we report only GrabCut results.”
Simple Does It: Weakly Supervised Instance and Semantic Segmentation (2016, methodology): “M ∩ G+ denotes the weakly or semi supervised model trained with MCG ∩ Grabcut+. Input image Ground truth Box Boxi Bbox-Seg+CRF MCG DenseCut GrabCut KGrabCut GrabCut+ GrabCut+i M ∩ G+ Input image Ground truth Box Boxi Bbox-Seg+CRF MCG DenseCut GrabCut KGrabCut GrabCut+ GrabCut+i M ∩ G+ Figure S2: Dif”
Interactive Full Image Segmentation by Considering All Regions Jointly (2018, background): “In contrast, in single object segmentation works [3, 8, 18, 16, 21, 24, 30, 31, 32, 38, 36, 37, 47, 60] both positive and negative annotations are made only on the target object and they are never shared, so they only have an effect on that one object.”
DeepCut: Object Segmentation from Bounding Box Annotations using Convolutional Neural Networks (2016, methodology): “In segmentation studies [1], [5], [10], bounding boxes are employed as both initialisation and spatial constraints for the segmentation problem.”