Multiscale Combinatorial Grouping

By Pablo Arbeláez, Jordi Pont-Tuset, Jonathan T. Barron, Ferran Marqués, Jitendra Malik. 2014

1,185 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

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation (2013, neutral): “Figure 3 compares R-CNN to the entries in the ILSVRC 2013 competition and to the post-competition OverFeat result [34].”
Mask R-CNN (2017, methodology): “DeepMask [32] and following works [33, 8] learn to propose segment candidates, which are then classified by Fast R-CNN.”
Recurrent Neural Networks for Semantic Instance Segmentation (2017, methodology): “[17] trained a two-branch network to simultaneously learn to classify bounding-box and segment proposals from [3].”
Fully Convolutional Instance-Aware Semantic Segmentation (2016, neutral): “Traditional approaches, e.g., MCG [1] and Selective Search [41], use low level image features.”
Recurrent Instance Segmentation (2015, background): “Indeed, they are the chosen model in most works reviewed in Sec.”
Hypercolumns for object segmentation and fine-grained localization (2014, methodology): “Training classifiers for segmentation and part localization: For each category we take bottom-up MCG candidates [1] that overlap a ground truth instance by 70% or more.”
Pixelwise Instance Segmentation with a Dynamically Instantiated Network (2017, methodology): “This system was based on the R-CNN pipeline [15]: Region proposals, generated by the method of [1], were classified into object categories with a Convolutional Neural Network (CNN) before applying bounding-box regression as post-processing.”
Shape-aware Instance Segmentation (2016, methodology): “Beyond instance-level semantic segmentation, many methods have been proposed to generate class-agnostic region proposals [1, 31, 18].”
Deep Watershed Transform for Instance Segmentation (2016, background): “For example, [1] generates object segments proposals, and reasons about combining them into object instances.”
InstanceCut: From Edges to Instances with MultiCut (2016, methodology): “A very similar approach generates proposals using e.g. MCG [3] and then, in the second stage, a different network classiﬁes these proposals [13, 21, 15, 11].”
Training Region-Based Object Detectors with Online Hard Example Mining (2016, methodology): “R-CNN, in contrast, uses region proposals [1, 2, 3, 4, 6, 10, 18, 32, 34], a method that was made popular by the selective search algorithm [32].”
The Cityscapes Dataset for Semantic Urban Scene Understanding (2016, methodology): “We train a Fast-R-CNN (FRCN) detector [20] on our training data in order to score MCG object proposals [1].”
Instance-Sensitive Fully Convolutional Networks (2016, methodology): “Previous instance semantic segmentation methods ( e.g. , [2,3,4,5]) in general resorted to oﬀ-the-shelf segment proposal methods ( e.g. , [6,7]).”
Instance-Aware Semantic Segmentation via Multi-task Network Cascades (2015, methodology): “nal features among masklevel proposals is enabled by using masking layers [7]. All these methods [10,13,14,7] rely on computationally expensive mask proposal methods. For example, the widely used MCG [1] takes 30 seconds processing an image, which becomes a bottleneck at inference time. DeepMask [25”
What Makes for Effective Detection Proposals? (2015, methodology): “Based on our results, we introduce a novel evaluation metric, the average recall (AR).”
Convolutional feature masking for joint object and stuff segmentation (2014, methodology): “This is 3.8% higher than the SDS result reported in [10] which uses AlexNet and MCG.”