Image Segmentation by Cascaded Region Agglomeration

By Zhile Ren, Gregory Shakhnarovich. 2013

123 citations recorded by the source corpus.

We propose a hierarchical segmentation algorithm that starts with a very fine over segmentation and gradually merges regions using a cascade of boundary classifiers. This approach allows the weights of region and boundary features to adapt to the segmentation scale at which they are applied. The stages of the cascade are trained sequentially, with asymetric loss to maximize boundary recall. On six segmentation data sets, our algorithm achieves best performance under most region-quality measures, and does it with fewer segments than the prior work. Our algorithm is also highly competitive in a dense over segmentation (super pixel) regime under boundary-based measures.

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

Fast Edge Detection Using Structured Forests (2014, critical): “Our full approach, SE+MS+SH, outperforms all state-ofthe-art approaches [1], [27], [31], [41], [42].”
Recurrent Pixel Embedding for Instance Grouping (2017, methodology): “Quantitatively, we compare our model to COB [64], HED [90], CEDN [91], LEP [66], UCM [1], ISCRA [75], NCuts [78], EGB [24], and the original mean shift (MShift) segmentation algorithm [15].”
SemiContour: A Semi-supervised Learning Approach for Contour Detection (2016, unknown): “….58 Felz-Hutt [15] .61 .64 .56 Normalized Cuts [10] .64 .68 .48 Mean Shift [9] .64 .68 .56 Gb [19] .69 .72 .72 gPb-owt-ucm [2] .73 .76 .70 ISCRA [28] - (.72) - (.75) - (.46) Sketch Tokens [20] .64(.73) .66(.75) .58(.78) SCG [35] .73(.74) .75(.76) .76(.77) SE [13] .66(.74) .68(.76) .69(.78)…”
Learning Deep Structured Multi-Scale Features using Attention-Gated CRFs for Contour Prediction (2017, methodology): “Felz-Hut [11], MeanShift [7], Normalized Cuts [32], ISCRA [30], gPb-ucm [1], SketchTokens [22], MCG [28], LEP [47], and more recent CNN-based methods, including DeepEdge [2], DeepCon-tour [31], HED [38], CEDN [42], COB [24].”
Multiscale Combinatorial Grouping (2014, methodology): “ISCRA [22] gPb-UCM [4] Quadtree MCG and CPMC”
Superpixels: An evaluation of the state-of-the-art (2016, unknown): “We refer to the corresponding publications: [75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 84, 88, 89].”
Instance Segmentation of Indoor Scenes Using a Coverage Loss (2014, methodology): “[21] and [10] both achieve the same weighted CUB scores but [21] is a bit more eﬃcient requiring fewer regions.”
Convolutional Oriented Boundaries: From Image Segmentation to High-Level Tasks (2016, unknown): “[29] 0.757 0.793 0.828 MCG [16] 0.747 0.779 0.759 UCM [15] 0.726 0.760 0.727 ISCRA [30] 0.724 0.752 0.783 NCuts [31] 0.641 0.674 0.447 EGB [32] 0.636 0.674 0.581 MShift [33] [16] 0.380 0.433 0.271 ISCRA [30] 0.352 0.418 0.275 UCM [15] 0.348 0.385 0.235 MShift [33] 0.229 0.292 0.122 NCuts [31] 0.213…”
Supervised Evaluation of Image Segmentation and Object Proposal Techniques (2016, methodology): “The state of the art of segmentation to compute the metameasures is represented in this paper by MCG-UCM [5], gPb-UCM [14], ISCRA [58], EGB [59], Normalized Cuts [60], Mean Shift [61], NWMC [62], IID-KL [63], and Saliency”
Semantic Image Segmentation with Contextual Hierarchical Models (2015, background): “Multi-scale processing of images has been proven critical in many computer vision tasks [8], [9].”
Object proposal by multi-branch hierarchical segmentation (2015, methodology): “The segmentation method in [24] also organizes the greedy merging process in cascaded stages with learned merging strategies, but based on different principles.”
SCALPEL: Segmentation Cascades with Localized Priors and Efficient Learning (2013, methodology): “Our cascade approach is most similar to that developed independently by [18], but we incorporate high-level information rather than taking a purely bottom-up approach.”
Richer Convolutional Features for Edge Detection (2016, methodology): “On the challenging BSDS500 dataset [9], we compare RCF with some well-known generic image segmenters, including NCut [42], MShift [43], EGB [44], gPb-UCM [9], ISCRA [45], MCG [6], LEP [46], and COB [40].”
Piecewise Flat Embedding for Image Segmentation (2018, methodology): “Starting from small superpixels, INSCRA [32] generates hierarchical segmentation with the help of a sequence of cascaded discriminative classifiers.”