Fast and flexible convolutional sparse coding

Heide, Felix; Heidrich, Wolfgang; Wetzstein, Gordon

doi:10.1109/cvpr.2015.7299149

Cited by 213 publications

(340 citation statements)

References 17 publications

Supporting

Mentioning

338

Contrasting

Order By: Relevance

“…Our work differs fundamentally from recent acceleration methods like those reported by Heide et al [31], Bristow et al [30], and Bao et al [33], [34], which rely on efficient mathematical formulations to solve the CSC optimisation problem. Such methods typically initialise filters with random values or by a discrete cosine transform (henceforth, DCT).…”

Section: Introduction and Related Workmentioning

confidence: 87%

“…CNN) tend to learn a subset of filters similar to well-known HCFs (e.g. Gabor filters, see [16], [20], [25], [26], [30], [31]). This is also the case for convolutional sparse coding (henceforth, CSC) as shown in Figure 1.…”

Section: Introduction and Related Workmentioning

confidence: 99%

“…[17], [32]), therefore our acceleration strategy, combined with state-of-the-art (and future) fast CSC solvers (e.g. [30], [31], [35]), could potentially contribute to advance the field further (e.g., faster retraining for different data sets and curvilinear structures, the possibility to learn a much larger filter bank which could lead to better segmentation performance, among others).…”

Section: Introduction and Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Accelerating Convolutional Sparse Coding for Curvilinear Structures Segmentation by Refining SCIRD-TS Filter Banks

Annunziata

Trucco

2016

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

Abstract-Deep learning has shown great potential for curvilinear structure (e.g. retinal blood vessels and neurites) segmentation as demonstrated by a recent auto-context regression architecture based on filter banks learned by convolutional sparse coding. However, learning such filter banks is very time-consuming, thus limiting the amount of filters employed and the adaptation to other data sets (i.e. slow re-training). We address this limitation by proposing a novel acceleration strategy to speedup convolutional sparse coding filter learning for curvilinear structure segmentation. Our approach is based on a novel initialisation strategy (warm start), and therefore it is different from recent methods improving the optimisation itself. Our warmstart strategy is based on carefully designed hand-crafted filters (SCIRD-TS), modelling appearance properties of curvilinear structures which are then refined by convolutional sparse coding. Experiments on four diverse data sets, including retinal blood vessels and neurites, suggest that the proposed method reduces significantly the time taken to learn convolutional filter banks (i.e. up to −82%) compared to conventional initialisation strategies. Remarkably, this speed-up does not worsen performance; in fact, filters learned with the proposed strategy often achieve a much lower reconstruction error and match or exceed the segmentation performance of random and DCT-based initialisation, when used as input to a random forest classifier.

show abstract

Section: Introduction and Related Workmentioning

confidence: 87%

Section: Introduction and Related Workmentioning

confidence: 99%

Section: Introduction and Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Accelerating Convolutional Sparse Coding for Curvilinear Structures Segmentation by Refining SCIRD-TS Filter Banks

Annunziata

Trucco

2016

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

show abstract

“…This approach is simpler than the boundary handling strategies described in [HHW15] and [Woh16a], and for many problems gives results of comparable quality. The functions defined here implement symmetric extension and cropping of images.…”

Section: Removal Of Impulse Noise Via Cscmentioning

confidence: 99%

SPORCO: A Python package for standard and convolutional sparse representations

Wohlberg¹

2017

Proceedings of the 16th Python in Science Conference

View full text Add to dashboard Cite

Abstract-SParse Optimization Research COde (SPORCO) is an open-sourcePython package for solving optimization problems with sparsity-inducing regularization, consisting primarily of sparse coding and dictionary learning, for both standard and convolutional forms of sparse representation. In the current version, all optimization problems are solved within the Alternating Direction Method of Multipliers (ADMM) framework. SPORCO was developed for applications in signal and image processing, but is also expected to be useful for problems in computer vision, statistics, and machine learning.Index Terms-sparse representations, convolutional sparse representations, sparse coding, convolutional sparse coding, dictionary learning, convolutional dictionary learning, alternating direction method of multipliers

show abstract

“…For the latter case, see [4,Section 3]; moreover, recent ADMM-based approaches compute the convolutions in the frequency domain [21], [22], [23], [4], [24].…”

Section: B Numerical Algorithms For (1) and (2)mentioning

confidence: 99%

A two-term penalty function for inverse problems with sparsity constrains

Rodríguez

2017

2017 25th European Signal Processing Conference (EUSIPCO)

View full text Add to dashboard Cite

Abstract-Inverse problems with sparsity constrains, such Basis Pursuit denoising (BPDN) and Convolutional BPDN (CBPDN), usually use the 1-norm as the penalty function; however such choice leads to a solution that is biased towards zero. Recently, several works have proposed and assessed the properties of other non-standard penalty functions (most of them non-convex), which avoid the above mentioned drawback and at the same time are intended to induce sparsity more strongly than the 1-norm.In this paper we propose a two-term penalty function consisting of a synthesis between the 1-norm and the penalty function associated with the Non-Negative Garrote (NNG) thresholding rule. Although the proposed two-term penalty function is nonconvex, the total cost function for the BPDN / CBPDN problems is still convex. The performance of the proposed twoterm penalty function is compared with other reported choices for practical denoising, deconvolution and convolutional sparse coding (CSC) problems within the BPDN / CBPDN frameworks. Our experimental results show that the proposed two-term penalty function is particularly effective (better reconstruction with sparser solutions) for the CSC problem while attaining competitive performance for the denoising and deconvolution problems.

show abstract

Fast and flexible convolutional sparse coding

Cited by 213 publications

References 17 publications

Accelerating Convolutional Sparse Coding for Curvilinear Structures Segmentation by Refining SCIRD-TS Filter Banks

Accelerating Convolutional Sparse Coding for Curvilinear Structures Segmentation by Refining SCIRD-TS Filter Banks

SPORCO: A Python package for standard and convolutional sparse representations

A two-term penalty function for inverse problems with sparsity constrains

Contact Info

Product

Resources

About