Manifold SLIC: A Fast Method to Compute Content-Sensitive Superpixels

Liu, Yong-Jin; Yu, Cuibo; Yu, Mingxin; He, Ying

doi:10.1109/cvpr.2016.77

Cited by 114 publications

(90 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Manifold SLIC [39] is extended from SLIC to compute content-sensitive superpixels, i.e. small superpixels in content-dense regions and large superpixels in content-sparse regions.…”

Section: Clustering Based Methodsmentioning

confidence: 99%

“…Popular clustering based methods include Simple Linear Iterative Clustering (SLIC) [39], Watershed [36], MeanShift [37], QuickShift [38], TurboPixel [10], Manifold SLIC [39], Contour Relaxed Superpixels (CRS) [8], Superpixel Segmentation using Linear Spectral Clustering (LSC) [4], Depth-Adaptive Superpixels (DAS) [18], Compact superpixels (CS) [12], Constant Intensity Superpixels (CIS) [12], SEEDS [17], VCells [19], VCCS [20], Superpixel Lattices [11], Lattices Cut [13],Scene Shape Superpixel (SSP) [40], Co-superpixel [41], Saliency-based superpixel (SBS) [5] and Structure Sensitive Superpixels (SSS) [9].…”

Section: Clustering Based Methodsmentioning

confidence: 99%

“…Therefore, 15 superpixel segmentation methods are chosen, which are SLIC [39], LSC [4], ERS [21], SEEDS [17], PB [14], CRS [8], LRW [7], TPS [16], CS [12][19], CIS [12], N-cut [3], Superpixel Lattices [11], VCells [19], Turbopixel [10], and Watershed [36]. Fig.1 shows the qualitative results of these methods.…”

Section: Evaluated Algorithmsmentioning

confidence: 99%

“…It is widely used in superpixel literature [1,18,19,39] to measure the segmentation quality. It measures the precision-recall curve which captures the trade-off between precision and accuracy.…”

Section: Precision-recallmentioning

confidence: 99%

See 3 more Smart Citations

Superpixel segmentation: A benchmark

Wang

Liu

Gao

et al. 2017

Signal Processing: Image Communication

119

View full text Add to dashboard Cite

“…Manifold SLIC [39] is extended from SLIC to compute content-sensitive superpixels, i.e. small superpixels in content-dense regions and large superpixels in content-sparse regions.…”

Section: Clustering Based Methodsmentioning

confidence: 99%

Section: Clustering Based Methodsmentioning

confidence: 99%

Section: Evaluated Algorithmsmentioning

confidence: 99%

Section: Precision-recallmentioning

confidence: 99%

See 2 more Smart Citations

Superpixel segmentation: A benchmark

Wang

Liu

Gao

et al. 2017

Signal Processing: Image Communication

119

View full text Add to dashboard Cite

“…Secondly, SLIC can generate superpixels with compact, regular size and shape. Moreover, SLIC is simple to use and understand [3,23]. Figure 1 displays the segmentation results of SEEDS, SuperPB, SLIC and Turbopixel.…”

Section: Introductionmentioning

confidence: 99%

BSLIC: SLIC Superpixels Based on Boundary Term

et al. 2017

View full text Add to dashboard Cite

A modified method for better superpixel generation based on simple linear iterative clustering (SLIC) is presented and named BSLIC in this paper. By initializing cluster centers in hexagon distribution and performing k-means clustering in a limited region, the generated superpixels are shaped into regular and compact hexagons. The additional cluster centers are initialized as edge pixels to improve boundary adherence, which is further promoted by incorporating the boundary term into the distance calculation of the k-means clustering. Berkeley Segmentation Dataset BSDS500 is used to qualitatively and quantitatively evaluate the proposed BSLIC method. Experimental results show that BSLIC achieves an excellent compromise between boundary adherence and regularity of size and shape. In comparison with SLIC, the boundary adherence of BSLIC is increased by at most 12.43% for boundary recall and 3.51% for under segmentation error.

show abstract

Superpixel Sampling Networks

Jampani

Sun

Liu

et al. 2018

Computer Vision – ECCV 2018

189

153

View full text Add to dashboard Cite

Superpixels provide an efficient low/mid-level representation of image data, which greatly reduces the number of image primitives for subsequent vision tasks. Existing superpixel algorithms are not differentiable, making them difficult to integrate into otherwise end-to-end trainable deep neural networks. We develop a new differentiable model for superpixel sampling that leverages deep networks for learning superpixel segmentation. The resulting Superpixel Sampling Network (SSN) is end-to-end trainable, which allows learning task-specific superpixels with flexible loss functions and has fast runtime. Extensive experimental analysis indicates that SSNs not only outperform existing superpixel algorithms on traditional segmentation benchmarks, but can also learn superpixels for other tasks. In addition, SSNs can be easily integrated into downstream deep networks resulting in performance improvements. Fig. 1:Overview of Superpixel Sampling Networks. A given image is first passed onto a deep network that extracts features at each pixel, which are then used by differentiable SLIC to generate the superpixels. Shown here are a couple of example SSN generated task-specific superpixels for semantic segmentation and optical flow.non-differentiable and thus using superpixels in deep networks introduces nondifferentiable modules in otherwise end-to-end trainable network architectures.In this work, we alleviate the second issue by proposing a new deep differentiable algorithm for superpixel segmentation. We start by revisiting the widely-used Simple Linear Iterative Clustering (SLIC) superpixel algorithm [1] and turn it into a differentiable algorithm by relaxing the nearest neighbor constraints present in SLIC. This new differentiable algorithm allows for end-to-end training and enables us to leverage powerful deep networks for learning superpixels instead of using traditional hand-crafted features. This combination of a deep network with differentiable SLIC forms our end-to-end trainable superpixel algorithm which we call Superpixel Sampling Network (SSN). Fig. 1 shows an overview of the proposed SSN. A given input image is first passed through a deep network producing features at each pixel. These deep features are then passed onto the differentiable SLIC, which performs iterative clustering, resulting in the desired superpixels. The entire network is end-to-end trainable. The differentiable nature of SSN allows the use of flexible loss functions for learning task-specific superpixels. Fig. 1 shows some sample SSN generated superpixels.Experimental results on 3 different segmentation benchmark datasets including BSDS500 [4], Cityscapes [10] and PascalVOC [11] indicate that the proposed superpixel sampling network (SSN) performs favourably against existing prominent superpixel algorithms, while also being faster. We also demonstrate that by simply integrating our SSN framework into an existing semantic segmentation network [13] that uses superpixels, performance improvements are achieved. In addition, we demonstrate the flex...

show abstract

Manifold SLIC: A Fast Method to Compute Content-Sensitive Superpixels

Cited by 114 publications

References 18 publications

Superpixel segmentation: A benchmark

Superpixel segmentation: A benchmark

BSLIC: SLIC Superpixels Based on Boundary Term

Superpixel Sampling Networks

Contact Info

Product

Resources

About