cvpr cvpr2013 cvpr2013-255 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Roberto Rigamonti, Amos Sironi, Vincent Lepetit, Pascal Fua
Abstract: Learning filters to produce sparse image representations in terms of overcomplete dictionaries has emerged as a powerful way to create image features for many different purposes. Unfortunately, these filters are usually both numerous and non-separable, making their use computationally expensive. In this paper, we show that such filters can be computed as linear combinations of a smaller number of separable ones, thus greatly reducing the computational complexity at no cost in terms of performance. This makes filter learning approaches practical even for large images or 3D volumes, and we show that we significantly outperform state-of-theart methods on the linear structure extraction task, in terms ofboth accuracy and speed. Moreover, our approach is general and can be used on generic filter banks to reduce the complexity of the convolutions.
Reference: text
sentIndex sentText sentNum sentScore
1 ch Abstract Learning filters to produce sparse image representations in terms of overcomplete dictionaries has emerged as a powerful way to create image features for many different purposes. [sent-3, score-0.69]
2 Unfortunately, these filters are usually both numerous and non-separable, making their use computationally expensive. [sent-4, score-0.56]
3 In this paper, we show that such filters can be computed as linear combinations of a smaller number of separable ones, thus greatly reducing the computational complexity at no cost in terms of performance. [sent-5, score-1.296]
4 This makes filter learning approaches practical even for large images or 3D volumes, and we show that we significantly outperform state-of-theart methods on the linear structure extraction task, in terms ofboth accuracy and speed. [sent-6, score-0.273]
5 Moreover, our approach is general and can be used on generic filter banks to reduce the complexity of the convolutions. [sent-7, score-0.322]
6 Introduction It has been shown that representing images as sparse lin- ear combinations of learned filters [27] yields effective approaches to image denoising and object recognition, which outperform those that rely on hand-crafted features [38]. [sent-9, score-0.828]
7 Unfortunately, because the filters are both numerous and not separable, they tend to be computationally demanding, which has slowed down their acceptance. [sent-11, score-0.56]
8 In this paper, we show that we can preserve the performance of these convolutional approaches while drastically reducing their cost by learning and using separable filters that approximate the non-separable ones. [sent-13, score-1.3]
9 Convolutional filter bank (a) learned for the extraction of linear structures in retinal scan images, along with its separable approximation (b). [sent-17, score-1.147]
10 The full-rank filters of (a) can be approximated very precisely as linear combinations of the far fewer separable filters of (b). [sent-18, score-1.836]
11 strates this behavior in the case of filters designed to classify whether or not a pixel belongs to a blood vessel in reti- nal scans. [sent-20, score-0.619]
12 Using the learned separable filters is much faster than using either the original non-separable ones or a stateof-the-art implementation of the FFT for all practical filter sizes. [sent-21, score-1.44]
13 1-based learning framework to directly learn a set of separable filters. [sent-24, score-0.599]
14 1(a), and then a second smaller set of separable filters, such as those of Fig. [sent-26, score-0.572]
15 Our contribution is therefore an original approach to approximating non-separable filters as a linear combination of a smaller set of separable ones. [sent-28, score-1.222]
16 It benefits both from the fact that there are fewer filters and that they are separable. [sent-29, score-0.588]
17 Furthermore, for the purpose of finding linear structures, our method is not only faster but also significantly more accurate than one of the best current techniques that relies on hand-designed filters [20]. [sent-30, score-0.682]
18 In the remainder of the paper, we first discuss related work, and then introduce our approach to separable approximation. [sent-31, score-0.572]
19 Recently, creating overcomplete dictionaries of features—sparse combinations of which can be used to represent images—has emerged as a powerful tool for object recognition [8, 18, 38] and image denoising [9, 24], among others. [sent-36, score-0.272]
20 This principle was exploited in [28] to avoid coarse discretizations of the scale and orientation spaces, yielding steerable separable 2D edge-detection kernels, and has been extended in [14]. [sent-39, score-0.645]
21 This precludes the arbitrary ones found in a learned dictionary, or the ones handcrafted to suit particular needs. [sent-41, score-0.191]
22 Two well known attempts to tackle the computational complexity issue by focusing on aspects other than separability are the steerable filters [12] and the gray-code filter kernels [4]. [sent-45, score-0.945]
23 An interesting recent attempt at reducing computational complexity is the approach of [32], which involves learning a filter bank by composing a few atoms from an handcrafted separable dictionary. [sent-51, score-1.084]
24 As shown in the results section, this results in a smaller number of separable filters that are tuned for the task at hand. [sent-53, score-1.132]
25 Formally, N filters {fj }1≤j≤N are computed as {fajr}g,{mminji}? [sent-58, score-0.56]
26 222777555533 with large amounts of data, as it is common in the medical domain, the required run-time convolutions are costly because the resulting filters are not separable. [sent-88, score-0.633]
27 uOldur d goal tios athe mreofroer me aton alogeokab floer O separable (fsil +ters t )w). [sent-92, score-0.572]
28 One way to do this would be to explicitly write the fj filters as products of 1D filters and to minimize the objective function of Eq. [sent-94, score-1.245]
29 (1), and directly forces the learned filters to be separable by lowering their rank. [sent-99, score-1.243]
30 Since arbitrary filters of rank R can be expressed as linear combinations of R separable filters [28], we replace fij the f filters of Eq. [sent-101, score-2.397]
31 (1) by linear combinations of filters that are forced to be separable by lowering their rank. [sent-102, score-1.293]
32 This solution is more general than the first, and retains the discriminative power of the full-rank filter bank. [sent-103, score-0.2]
33 Penalizing High-Rank Filters A straightforward approach to finding low-rank filters is to add a penalty term to the objective function of Eq. [sent-106, score-0.586]
34 The nuclear norm of a matrix is the sum of its singular values and is a convex relaxation of the rank [11]. [sent-136, score-0.22]
35 Thus, forcing the nuclear norm to be small amounts to lowering the rank of the filters. [sent-137, score-0.23]
36 Experimentally, for sufficiently high values of λ∗, the sj filters become effectively rank 1 and can be written as products of 1D filters. [sent-138, score-0.62]
37 (2), which has the nuclear norm of the filters as an additional term compared to Eq. [sent-140, score-0.693]
38 After taking a step in the direction opposite of that of the gradient of the filters, as described in the previous section, we just have to apply the proximal operator of ∗ the nuclear norm to the filters. [sent-142, score-0.188]
39 on each filter s, soft-thresholding the values of the diagonal matrix D to obtain a new matrix D? [sent-144, score-0.179]
40 We have found it effective to start with a low value of λ∗, solve the system, and then progressively increase it until the filter ranks are close to one. [sent-155, score-0.179]
41 Linear Combinations of Separable Filters In this second approach, we write the N fj filters of Eq. [sent-158, score-0.628]
42 (1) as linear combinations of M separable filters {sk}1≤k≤M. [sent-159, score-1.248]
43 nts to convolving it with the separable sk filters and then ? [sent-163, score-1.223]
44 Again, we introduce the nuclear norm to force the sk filters to be separable. [sent-185, score-0.737]
45 We tried instead a simpler approach, which has yielded better results by decoupling the computation of the non-separable filters from that of the separable ones. [sent-190, score-1.132]
46 We first learn a set of nonseparable filters {fj } by optimizing the original objective fsuepncatriaobnl eo ffi Elteq. [sent-191, score-0.63]
47 f W}e b tyhe onp tloimokiz finorg ste hpear oarbiglei nfial te orsb wjechtoivsee linear combinations approximate the fj filters by solving wkj {sark}gm,{wiknj}? [sent-193, score-0.838]
48 Examples of non-separable and separable 3D filter banks, learned on the OPF dataset [2]. [sent-212, score-0.817]
49 Response of the classifier trained on the separable filter bank output (b) (d). [sent-214, score-0.944]
50 (d) The separable filter bank learned by optimizing Eq. [sent-217, score-0.987]
51 Fortunately, our approach generalizes naturally to learning 3D separable filters. [sent-222, score-0.599]
52 It simply involves a CPD of the filters with R set to a large enough value, followed by a soft-thresholding on their coefficients σrk. [sent-232, score-0.583]
53 2(d) depicts an example of the 3D separable filter we obtained and can be compared to the nonseparable ones of Fig. [sent-236, score-0.825]
54 Results and Discussion To demonstrate our approach on both 2D and 3D data, we first compare the performance of our separable filters against that of non-separable ones for the purpose of classifying pixels and voxels in biomedical images as belonging to linear structures or not. [sent-240, score-1.352]
55 We show that our separable filters systematically deliver a very substantial speed-up at no loss in performance. [sent-241, score-1.153]
56 We then demonstrate that they can also approximate very effectively non-separable ones learned for denoising purposes. [sent-242, score-0.214]
57 For the purpose ofthese comparisons, we will refer to the non-separable filters obtained by minimizing the objective function of Eq. [sent-243, score-0.614]
58 (1) as NON-SEP, and the separable ones learned using the technique of Sections 3. [sent-244, score-0.668]
59 We will denote by SEP-SVD the separable filters obtained by approximating each NON-SEP filter by the outer product of its first left singular vector with its first right singular vector, which is the simplest way to approximate a non-separable filter by a separable one. [sent-247, score-2.262]
60 2, linear combinations of the SEP-COMB filters can be used to represent the NON-SEP ones. [sent-249, score-0.676]
61 This approach, which we will refer to as SEP-COMB∗, further simplifies the run-time computations because the linear combinations are implicitly learned by the classifier at training-time. [sent-251, score-0.225]
62 Here, we demonstrate the power of our separable filters for the purpose of identifying linear structures, a long-standing Computer Vision problem that still remains wide-open when the image data is noisy. [sent-255, score-1.215]
63 In [3 1], it was shown that convolving images with non-separable filter banks learned by 222777555755 solving the problem of Eq. [sent-259, score-0.412]
64 (1) and training an SVM on the output of those filters outperforms these other methods. [sent-260, score-0.56]
65 Note that we do not need to compute the linear combination of the filter outputs in the case of SEP-COMB, since the Random Forest classifier relies on linear projections. [sent-277, score-0.297]
66 In particular, the graph for the 2D case reports the time in second needed to convolve a 512 512 2D image with a bank noef 1d2e1d fi tolte crso,n as a efu anc 5ti1o2n × ×of 5 1th2e f2iDlte rim msiazgee, by using atnhek MATLAB’s conv2 function, the FFTW library, and our method 2. [sent-286, score-0.199]
67 They are constant for FFTW but much higher even for relatively large filters and even if the image size is optimal for FFTW as it is a power of 2. [sent-289, score-0.581]
68 ×× When a 128 128 64 3D volume is considered, the advantage onf a our m×e1t2ho8d× i6s even vcolleuamreer, as tohnes icduebriecd complexity makes the computations in the non-separable case impractical even when very few filters are considered. [sent-290, score-0.625]
69 Indeed, the reduction in computational time is largely due to the separability of the filters, and only partially to the reduction of the numbers of the filters involved. [sent-291, score-0.647]
70 We first learned a filter bank with 121 learned filters of size 21 21 on the DRIVE dataset and one on the BF2D sdiazteas 2e1t, as 1the osne parameters provided us owneith o nth teh ebe BstF results. [sent-293, score-1.041]
71 To assess how well our approach generalized, we also used the filter bank learned for the DRIVE dataset for the STARE dataset. [sent-294, score-0.415]
72 We have then learned other filter banks of reduced cardinality, both full-rank and separable, to assess the impact of the filter bank size on the final classification performance. [sent-296, score-0.736]
73 The costs of the Fourier Transform are indeed amortised only for extremely large image and filter sizes. [sent-309, score-0.201]
74 We learned the 3D filter bank made of 49 13 13 13 pixel afirlnteersd depicted by Fig. [sent-317, score-0.415]
75 2k(c m) aadned othfe 4 196 1 separable f 1il3ters of Fig. [sent-318, score-0.572]
76 Since these classifiers do not require us to compute the linear combination of the separable filter outputs, we chose again the SEP-COMB∗ approach for our experiments. [sent-324, score-0.785]
77 We compare SEP-COMB∗ against NON-SEP-FFT, a Fourier-based implementation of NON-SEP, a 3D version of OOF, and SEP-CPD, which approximates each filter by its rank-one CPD decomposition and is therefore a 3D equivalent of SEP-SVD. [sent-328, score-0.213]
78 Denoising To evaluate how well our SEP-COMB approach is at representing a set of generic filters in a very different context, we used it to approximate the 256 denoising filters computed by the K-SVD algorithm [9], some of which are depicted by Fig. [sent-336, score-1.238]
79 We experimented with different sizes of the approximating separable filter bank, and reported the results in Tab. [sent-338, score-0.807]
80 As can be seen, the 36 separable filters shown in Fig. [sent-340, score-1.132]
81 6(a) are already enough to obtain a very accurate approximation, giving a perfect reconstruction of the original filters up to a nearly imperceptible smoothing of the filters with many high-frequency components. [sent-341, score-1.12]
82 [32] also considered the approximation of filter banks learned with the K-SVD algorithm by using sparse linear combinations 222777555977 Figure 4. [sent-344, score-0.511]
83 The graphs compare the F-measure [36] obtained by the different approaches, normalizing the result by the F-measure obtained with the Optimally Oriented Flux filter [20]. [sent-346, score-0.179]
84 Our filtering approach outperforms the OOF results in all the datasets, and the separable filters do so at a fraction of the computational costs of the non-separable filters, while retaining their accuracy. [sent-347, score-1.179]
85 Times are given in seconds and represent the time it takes to convolve the input image with the considered filter banks. [sent-348, score-0.208]
86 The F-measure is normalized by the F-measure obtained with the Optimally Oriented Flux filter [20]. [sent-352, score-0.179]
87 However, we need significantly fewer separable filters, only 36 compared to the 100 for [32]. [sent-355, score-0.6]
88 Interestingly, the basis of separable filters we learn seem general. [sent-356, score-1.132]
89 We proved that by taking the filters that were learned to approximate a filter bank of a specific image, and we used them to reconstruct the filter banks of the other im- ages. [sent-357, score-1.302]
90 In other words, we kept the same sk filters learned for the Barbara image, and only optimized on the weights in Eq. [sent-358, score-0.67]
91 (a) The 36 separable filters learned by SEP-COMB to approximate a bank of 256 filters learned by K-SVD algorithm of [9]. [sent-364, score-2.022]
92 (b) Comparison between some of the original filters learned by K-SVD (top row) and their approximations reconstructed by our algorithm (bottom row). [sent-365, score-0.626]
93 While filters with a regular structure are very well approximated, noisy filters are slightly smoothed by the approximation. [sent-366, score-1.12]
94 Conclusion We have proposed a learning-based filtering scheme applied to the extraction of linear structures, along with two learning-based strategies for obtaining separable filter banks. [sent-369, score-0.818]
95 The first one directly learns separable filters by modifying the regular objective function. [sent-370, score-1.158]
96 The second one learns a basis of separable filters to approximate an existing filter bank, and not only gets the same performance of the original, but also considerably reduces the number of filters, and thus convolutions, required. [sent-371, score-1.339]
97 The images were first artificially corrupted by additive Gaussian white noise with standard deviation 20, and denoised with the K-SVD method [9], using the bank of 256 filters computed by the original method and its approximations we obtained with our SEP-COMB methods. [sent-376, score-0.73]
98 SEP-COMB-Barbara denotes the strategy where, instead of grounding the reconstruction on the approximating filter bank corresponding to the image to denoise, the approximating filter bank from the Barbara image is used. [sent-378, score-0.81]
99 This filter bank seems general as it does not degrade the results. [sent-379, score-0.349]
100 Moreover, designers of handcrafted filter banks do not have to restrict themselves to separable filters anymore: they can freely choose filters for the application at hand, and approximate them using few separable filters with our approach. [sent-417, score-3.216]
wordName wordTfidf (topN-words)
[('separable', 0.572), ('filters', 0.56), ('filter', 0.179), ('bank', 0.17), ('banks', 0.12), ('convolutional', 0.113), ('mij', 0.11), ('nuclear', 0.094), ('denoising', 0.09), ('oof', 0.087), ('combinations', 0.082), ('steerable', 0.073), ('fj', 0.068), ('rigamonti', 0.066), ('wkj', 0.066), ('learned', 0.066), ('handcrafted', 0.065), ('separability', 0.062), ('biomedical', 0.062), ('fftw', 0.059), ('singular', 0.058), ('approximating', 0.056), ('barbara', 0.054), ('stacks', 0.052), ('tensor', 0.051), ('drive', 0.05), ('convolutions', 0.05), ('cpd', 0.049), ('retinal', 0.049), ('flux', 0.047), ('convolving', 0.047), ('lowering', 0.045), ('structures', 0.044), ('nonseparable', 0.044), ('stare', 0.044), ('sk', 0.044), ('opf', 0.041), ('diadem', 0.039), ('norm', 0.039), ('volumes', 0.039), ('emerged', 0.038), ('optimally', 0.037), ('unfortunately', 0.036), ('epfl', 0.036), ('lecun', 0.035), ('decomposition', 0.034), ('linear', 0.034), ('proximal', 0.034), ('faster', 0.033), ('extraction', 0.033), ('overcomplete', 0.032), ('vessel', 0.031), ('mairal', 0.031), ('products', 0.031), ('sparse', 0.03), ('olshausen', 0.03), ('cvlab', 0.03), ('dictionaries', 0.03), ('ones', 0.03), ('convolve', 0.029), ('lausanne', 0.029), ('tmi', 0.029), ('rank', 0.029), ('supplementary', 0.028), ('bottou', 0.028), ('blood', 0.028), ('fourier', 0.028), ('convolution', 0.028), ('fewer', 0.028), ('purpose', 0.028), ('approximate', 0.028), ('learning', 0.027), ('relies', 0.027), ('objective', 0.026), ('computational', 0.025), ('material', 0.025), ('latter', 0.024), ('psnr', 0.023), ('classifier', 0.023), ('fua', 0.023), ('complexity', 0.023), ('involves', 0.023), ('amounts', 0.023), ('kernels', 0.023), ('unknowns', 0.022), ('sapiro', 0.022), ('voxels', 0.022), ('classification', 0.022), ('impractical', 0.022), ('hardware', 0.022), ('costs', 0.022), ('systematically', 0.021), ('supplemental', 0.021), ('networks', 0.021), ('power', 0.021), ('voxel', 0.021), ('operator', 0.021), ('svd', 0.021), ('computations', 0.02), ('roc', 0.02)]
simIndex simValue paperId paperTitle
same-paper 1 1.0000005 255 cvpr-2013-Learning Separable Filters
Author: Roberto Rigamonti, Amos Sironi, Vincent Lepetit, Pascal Fua
Abstract: Learning filters to produce sparse image representations in terms of overcomplete dictionaries has emerged as a powerful way to create image features for many different purposes. Unfortunately, these filters are usually both numerous and non-separable, making their use computationally expensive. In this paper, we show that such filters can be computed as linear combinations of a smaller number of separable ones, thus greatly reducing the computational complexity at no cost in terms of performance. This makes filter learning approaches practical even for large images or 3D volumes, and we show that we significantly outperform state-of-theart methods on the linear structure extraction task, in terms ofboth accuracy and speed. Moreover, our approach is general and can be used on generic filter banks to reduce the complexity of the convolutions.
2 0.22611804 392 cvpr-2013-Separable Dictionary Learning
Author: Simon Hawe, Matthias Seibert, Martin Kleinsteuber
Abstract: Many techniques in computer vision, machine learning, and statistics rely on the fact that a signal of interest admits a sparse representation over some dictionary. Dictionaries are either available analytically, or can be learned from a suitable training set. While analytic dictionaries permit to capture the global structure of a signal and allow a fast implementation, learned dictionaries often perform better in applications as they are more adapted to the considered class of signals. In imagery, unfortunately, the numerical burden for (i) learning a dictionary and for (ii) employing the dictionary for reconstruction tasks only allows to deal with relatively small image patches that only capture local image information. The approach presented in this paper aims at overcoming these drawbacks by allowing a separable structure on the dictionary throughout the learning process. On the one hand, this permits larger patch-sizes for the learning phase, on the other hand, the dictionary is applied efficiently in reconstruction tasks. The learning procedure is based on optimizing over a product of spheres which updates the dictionary as a whole, thus enforces basic dictionary proper- , ties such as mutual coherence explicitly during the learning procedure. In the special case where no separable structure is enforced, our method competes with state-of-the-art dictionary learning methods like K-SVD.
3 0.21111715 346 cvpr-2013-Real-Time No-Reference Image Quality Assessment Based on Filter Learning
Author: Peng Ye, Jayant Kumar, Le Kang, David Doermann
Abstract: This paper addresses the problem of general-purpose No-Reference Image Quality Assessment (NR-IQA) with the goal ofdeveloping a real-time, cross-domain model that can predict the quality of distorted images without prior knowledge of non-distorted reference images and types of distortions present in these images. The contributions of our work are two-fold: first, the proposed method is highly efficient. NR-IQA measures are often used in real-time imaging or communication systems, therefore it is important to have a fast NR-IQA algorithm that can be used in these real-time applications. Second, the proposed method has the potential to be used in multiple image domains. Previous work on NR-IQA focus primarily on predicting quality of natural scene image with respect to human perception, yet, in other image domains, the final receiver of a digital image may not be a human. The proposed method consists of the following components: (1) a local feature extractor; (2) a global feature extractor and (3) a regression model. While previous approaches usually treat local feature extraction and regres- sion model training independently, we propose a supervised method based on back-projection, which links the two steps by learning a compact set of filters which can be applied to local image patches to obtain discriminative local features. Using a small set of filters, the proposed method is extremely fast. We have tested this method on various natural scene and document image datasets and obtained stateof-the-art results.
4 0.20096718 163 cvpr-2013-Fast, Accurate Detection of 100,000 Object Classes on a Single Machine
Author: Thomas Dean, Mark A. Ruzon, Mark Segal, Jonathon Shlens, Sudheendra Vijayanarasimhan, Jay Yagnik
Abstract: Many object detection systems are constrained by the time required to convolve a target image with a bank of filters that code for different aspects of an object’s appearance, such as the presence of component parts. We exploit locality-sensitive hashing to replace the dot-product kernel operator in the convolution with a fixed number of hash-table probes that effectively sample all of the filter responses in time independent of the size of the filter bank. To show the effectiveness of the technique, we apply it to evaluate 100,000 deformable-part models requiring over a million (part) filters on multiple scales of a target image in less than 20 seconds using a single multi-core processor with 20GB of RAM. This represents a speed-up of approximately 20,000 times— four orders of magnitude— when compared withperforming the convolutions explicitly on the same hardware. While mean average precision over the full set of 100,000 object classes is around 0.16 due in large part to the challenges in gathering training data and collecting ground truth for so many classes, we achieve a mAP of at least 0.20 on a third of the classes and 0.30 or better on about 20% of the classes.
Author: Chao Liu, Geifei Yang, Jinwei Gu
Abstract: We present a computational imaging method for raw material classification using features of Bidirectional Texture Functions (BTF). Texture is an intrinsic feature for many materials, such as wood, fabric, and granite. At appropriate scales, even “uniform” materials will also exhibit texture features that can be helpful for recognition, such as paper, metal, and ceramic. To cope with the high-dimensionality of BTFs, in this paper, we proposed to learn discriminative illumination patterns and texture filters, with which we can directly measure optimal projections of BTFs for classification. We also studied the effects of texture rotation and scale variation for material classification. We built an LED-based multispectral dome, with which we have acquired a BTF database of a variety of materials and demonstrated the effectiveness of the proposed approach for material classification.
6 0.19011118 164 cvpr-2013-Fast Convolutional Sparse Coding
7 0.14826721 388 cvpr-2013-Semi-supervised Learning of Feature Hierarchies for Object Detection in a Video
8 0.11400077 165 cvpr-2013-Fast Energy Minimization Using Learned State Filters
9 0.11304732 328 cvpr-2013-Pedestrian Detection with Unsupervised Multi-stage Feature Learning
10 0.11221683 369 cvpr-2013-Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination
11 0.11079059 205 cvpr-2013-Hollywood 3D: Recognizing Actions in 3D Natural Scenes
12 0.1107472 104 cvpr-2013-Deep Convolutional Network Cascade for Facial Point Detection
13 0.10958868 198 cvpr-2013-Handling Noise in Single Image Deblurring Using Directional Filters
14 0.084045924 169 cvpr-2013-Fast Patch-Based Denoising Using Approximated Patch Geodesic Paths
15 0.07780239 350 cvpr-2013-Reconstructing Loopy Curvilinear Structures Using Integer Programming
16 0.068151802 204 cvpr-2013-Histograms of Sparse Codes for Object Detection
17 0.067671627 131 cvpr-2013-Discriminative Non-blind Deblurring
18 0.066674903 405 cvpr-2013-Sparse Subspace Denoising for Image Manifolds
19 0.066504069 378 cvpr-2013-Sampling Strategies for Real-Time Action Recognition
20 0.064825103 427 cvpr-2013-Texture Enhanced Image Denoising via Gradient Histogram Preservation
topicId topicWeight
[(0, 0.15), (1, -0.021), (2, -0.062), (3, 0.09), (4, -0.008), (5, 0.039), (6, 0.034), (7, 0.004), (8, -0.05), (9, -0.037), (10, -0.048), (11, -0.035), (12, -0.005), (13, -0.079), (14, 0.029), (15, 0.059), (16, 0.002), (17, 0.004), (18, 0.147), (19, 0.069), (20, 0.006), (21, 0.018), (22, -0.019), (23, -0.023), (24, -0.026), (25, 0.095), (26, -0.035), (27, -0.035), (28, -0.036), (29, -0.011), (30, -0.126), (31, -0.062), (32, -0.002), (33, -0.129), (34, -0.068), (35, 0.156), (36, 0.042), (37, 0.075), (38, -0.017), (39, 0.14), (40, -0.063), (41, -0.054), (42, 0.016), (43, -0.094), (44, -0.188), (45, -0.064), (46, 0.07), (47, -0.02), (48, 0.01), (49, 0.01)]
simIndex simValue paperId paperTitle
same-paper 1 0.96624684 255 cvpr-2013-Learning Separable Filters
Author: Roberto Rigamonti, Amos Sironi, Vincent Lepetit, Pascal Fua
Abstract: Learning filters to produce sparse image representations in terms of overcomplete dictionaries has emerged as a powerful way to create image features for many different purposes. Unfortunately, these filters are usually both numerous and non-separable, making their use computationally expensive. In this paper, we show that such filters can be computed as linear combinations of a smaller number of separable ones, thus greatly reducing the computational complexity at no cost in terms of performance. This makes filter learning approaches practical even for large images or 3D volumes, and we show that we significantly outperform state-of-theart methods on the linear structure extraction task, in terms ofboth accuracy and speed. Moreover, our approach is general and can be used on generic filter banks to reduce the complexity of the convolutions.
2 0.75704461 164 cvpr-2013-Fast Convolutional Sparse Coding
Author: Hilton Bristow, Anders Eriksson, Simon Lucey
Abstract: Sparse coding has become an increasingly popular method in learning and vision for a variety of classification, reconstruction and coding tasks. The canonical approach intrinsically assumes independence between observations during learning. For many natural signals however, sparse coding is applied to sub-elements (i.e. patches) of the signal, where such an assumption is invalid. Convolutional sparse coding explicitly models local interactions through the convolution operator, however the resulting optimization problem is considerably more complex than traditional sparse coding. In this paper, we draw upon ideas from signal processing and Augmented Lagrange Methods (ALMs) to produce a fast algorithm with globally optimal subproblems and super-linear convergence.
3 0.75658154 346 cvpr-2013-Real-Time No-Reference Image Quality Assessment Based on Filter Learning
Author: Peng Ye, Jayant Kumar, Le Kang, David Doermann
Abstract: This paper addresses the problem of general-purpose No-Reference Image Quality Assessment (NR-IQA) with the goal ofdeveloping a real-time, cross-domain model that can predict the quality of distorted images without prior knowledge of non-distorted reference images and types of distortions present in these images. The contributions of our work are two-fold: first, the proposed method is highly efficient. NR-IQA measures are often used in real-time imaging or communication systems, therefore it is important to have a fast NR-IQA algorithm that can be used in these real-time applications. Second, the proposed method has the potential to be used in multiple image domains. Previous work on NR-IQA focus primarily on predicting quality of natural scene image with respect to human perception, yet, in other image domains, the final receiver of a digital image may not be a human. The proposed method consists of the following components: (1) a local feature extractor; (2) a global feature extractor and (3) a regression model. While previous approaches usually treat local feature extraction and regres- sion model training independently, we propose a supervised method based on back-projection, which links the two steps by learning a compact set of filters which can be applied to local image patches to obtain discriminative local features. Using a small set of filters, the proposed method is extremely fast. We have tested this method on various natural scene and document image datasets and obtained stateof-the-art results.
Author: Chao Liu, Geifei Yang, Jinwei Gu
Abstract: We present a computational imaging method for raw material classification using features of Bidirectional Texture Functions (BTF). Texture is an intrinsic feature for many materials, such as wood, fabric, and granite. At appropriate scales, even “uniform” materials will also exhibit texture features that can be helpful for recognition, such as paper, metal, and ceramic. To cope with the high-dimensionality of BTFs, in this paper, we proposed to learn discriminative illumination patterns and texture filters, with which we can directly measure optimal projections of BTFs for classification. We also studied the effects of texture rotation and scale variation for material classification. We built an LED-based multispectral dome, with which we have acquired a BTF database of a variety of materials and demonstrated the effectiveness of the proposed approach for material classification.
5 0.66599029 369 cvpr-2013-Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination
Author: Laurent Sifre, Stéphane Mallat
Abstract: An affine invariant representation is constructed with a cascade of invariants, which preserves information for classification. A joint translation and rotation invariant representation of image patches is calculated with a scattering transform. It is implemented with a deep convolution network, which computes successive wavelet transforms and modulus non-linearities. Invariants to scaling, shearing and small deformations are calculated with linear operators in the scattering domain. State-of-the-art classification results are obtained over texture databases with uncontrolled viewing conditions.
6 0.55466461 163 cvpr-2013-Fast, Accurate Detection of 100,000 Object Classes on a Single Machine
7 0.52541137 328 cvpr-2013-Pedestrian Detection with Unsupervised Multi-stage Feature Learning
8 0.52530974 304 cvpr-2013-Multipath Sparse Coding Using Hierarchical Matching Pursuit
9 0.50795221 388 cvpr-2013-Semi-supervised Learning of Feature Hierarchies for Object Detection in a Video
10 0.50559986 350 cvpr-2013-Reconstructing Loopy Curvilinear Structures Using Integer Programming
11 0.49962142 104 cvpr-2013-Deep Convolutional Network Cascade for Facial Point Detection
12 0.49535415 204 cvpr-2013-Histograms of Sparse Codes for Object Detection
13 0.48685986 427 cvpr-2013-Texture Enhanced Image Denoising via Gradient Histogram Preservation
14 0.48633152 83 cvpr-2013-Classification of Tumor Histology via Morphometric Context
15 0.45998329 421 cvpr-2013-Supervised Kernel Descriptors for Visual Recognition
16 0.45352992 169 cvpr-2013-Fast Patch-Based Denoising Using Approximated Patch Geodesic Paths
17 0.44349948 35 cvpr-2013-Adaptive Compressed Tomography Sensing
18 0.44048482 64 cvpr-2013-Blessing of Dimensionality: High-Dimensional Feature and Its Efficient Compression for Face Verification
19 0.43795058 351 cvpr-2013-Recovering Line-Networks in Images by Junction-Point Processes
20 0.42436993 266 cvpr-2013-Learning without Human Scores for Blind Image Quality Assessment
topicId topicWeight
[(10, 0.116), (16, 0.03), (26, 0.051), (28, 0.022), (33, 0.297), (38, 0.011), (57, 0.197), (67, 0.048), (69, 0.072), (80, 0.011), (87, 0.054)]
simIndex simValue paperId paperTitle
1 0.94261366 141 cvpr-2013-Efficient Computation of Shortest Path-Concavity for 3D Meshes
Author: Henrik Zimmer, Marcel Campen, Leif Kobbelt
Abstract: In the context of shape segmentation and retrieval object-wide distributions of measures are needed to accurately evaluate and compare local regions ofshapes. Lien et al. [16] proposed two point-wise concavity measures in the context of Approximate Convex Decompositions of polygons measuring the distance from a point to the polygon ’s convex hull: an accurate Shortest Path-Concavity (SPC) measure and a Straight Line-Concavity (SLC) approximation of the same. While both are practicable on 2D shapes, the exponential costs of SPC in 3D makes it inhibitively expensive for a generalization to meshes [14]. In this paper we propose an efficient and straight forward approximation of the Shortest Path-Concavity measure to 3D meshes. Our approximation is based on discretizing the space between mesh and convex hull, thereby reducing the continuous Shortest Path search to an efficiently solvable graph problem. Our approach works outof-the-box on complex mesh topologies and requires no complicated handling of genus. Besides presenting a rigorous evaluation of our method on a variety of input meshes, we also define an SPC-based Shape Descriptor and show its superior retrieval and runtime performance compared with the recently presented results on the Convexity Distribution by Lian et al. [12].
2 0.90702051 95 cvpr-2013-Continuous Inference in Graphical Models with Polynomial Energies
Author: Mathieu Salzmann
Abstract: In this paper, we tackle the problem of performing inference in graphical models whose energy is a polynomial function of continuous variables. Our energy minimization method follows a dual decomposition approach, where the global problem is split into subproblems defined over the graph cliques. The optimal solution to these subproblems is obtained by making use of a polynomial system solver. Our algorithm inherits the convergence guarantees of dual decomposition. To speed up optimization, we also introduce a variant of this algorithm based on the augmented Lagrangian method. Our experiments illustrate the diversity of computer vision problems that can be expressed with polynomial energies, and demonstrate the benefits of our approach over existing continuous inference methods.
3 0.90009314 79 cvpr-2013-Cartesian K-Means
Author: Mohammad Norouzi, David J. Fleet
Abstract: A fundamental limitation of quantization techniques like the k-means clustering algorithm is the storage and runtime cost associated with the large numbers of clusters required to keep quantization errors small and model fidelity high. We develop new models with a compositional parameterization of cluster centers, so representational capacity increases super-linearly in the number of parameters. This allows one to effectively quantize data using billions or trillions of centers. We formulate two such models, Orthogonal k-means and Cartesian k-means. They are closely related to one another, to k-means, to methods for binary hash function optimization like ITQ [5], and to Product Quantization for vector quantization [7]. The models are tested on largescale ANN retrieval tasks (1M GIST, 1B SIFT features), and on codebook learning for object recognition (CIFAR-10). 1. Introduction and background Techniques for vector quantization, like the well-known k-means algorithm, are used widely in vision and learning. Common applications include codebook learning for object recognition [16], and approximate nearest neighbor (ANN) search for information retrieval [12, 14]. In general terms, such techniques involve partitioning an input vector space into multiple regions {Si}ik=1, mapping points x in each region uinlttiop region-specific representatives {ci}ik=1, known as cioennte irnst.o A resg siuonch-,s a quantizer, qse(nxt)a,t can b {ec expressed as k q(x) = ?1(x∈Si)ci, (1) i?= ?1 where 1(·) is the usual indicator function. eTrhee 1 quality oef u a quantizer oisr expressed in terms of expected distortion, a common measure of which is squared error ?x − q(x) ?22. In this case, given centers {ci}, the region t ?ox xw −hic qh(x a point nis t assigned gwivitehn m ceinnitemrasl {dcis}to,r tthioen r eisobtained by Euclidean nearest neighbor (NN) search. The k-means algorithm can be used to learn centers from data. To reduce expected distortion, crucial for many applications, one can shrink region volumes by increasing k, the number of regions. In practice, however, increasing k results in prohibitive storage and run-time costs. Even if one resorts to ANN search with approximate k-means [14] or hierarchical k-means [12], it is hard to scale to large k (e.g., k = 264), as storing the centers is untenable. This paper concerns methods for scalable quantization with tractable storage and run-time costs. Inspired by Product Quantization (PQ), a state-of-the-art algorithm for ANN search with high-dimensional data (e.g., [7]), compositionality is one key. By expressing data in terms of recurring, reusable parts, the representational capacity of compositional models grows exponentially in the number ofparameters. Compression techniques like JPEG accomplish this by encoding images as disjoint rectangular patches. PQ divides the feature space into disjoint subspaces that are quantized independently. Other examples include part-based recognition models (e.g., [18]), and tensor-based models for stylecontent separation (e.g., [17]). Here, with a compositional parameterization of region centers, we find a family of models that reduce the enc√oding cost of k centers down from k to between log2 k and √k. A model parameter controls the trade-off between model fidelity and compactness. We formulate two related algorithms, Orthogonal kmeans (ok-means) and Cartesian k-means (ck-means). They are natural extensions of k-means, and are closely related to other hashing and quantization methods. The okmeans algorithm is a generalization of the Iterative Quantization (ITQ) algorithm for finding locality-sensitive binary codes [5]. The ck-means model is an extension of okmeans, and can be viewed as a generalization of PQ.1 We then report empirical results on large-scale ANN search tasks, and on codebook learning for recognition. For ANN search we use datasets of 1M GIST and 1B SIFT features, with both symmetric and asymmetric distance measures on the latent codes. We consider codebook learning for a generic form of recognition on CIFAR-10. 2. k-means Given a dataset of n p-dim points, D ≡ {xj }jn=1, the kmeans algorithm partitions ithme n points in ≡to { kx c}lusters, and 1A very similar generalization of PQ, was developed concurrently by Ge, He, Ke and Sun, and also appears in CVPR 2013 [4]. 333000111755 represents each cluster by a center point. Let C ∈ Rp×k breep a mseantrtsix e wachhos celu sctoelurm byns a comprise tihnet .k Lceltus Cter ∈ centers, i.e., C = [c1, c2 , · · · , ck] . The k-means objective is to minimize the within,-c··lu·s ,tecr squared distances: ?k-means(C) = = x?∈Dmiin?x − ci?22 x?∈Db∈mHin1/k?x − Cb?22 (2) where H1/k ≡ {b | b ∈ {0, 1}k and ?b? = 1}, i.e., b is a binary vee Hctor comprising a ,1-1o}f-ka encoding. Lloyd’s b k- imse aa bni-s algorithm [11] finds a local minimum of (2) by iterative, alternating optimization with respect to C and the b’s. The k-means model is simple and intuitive, using NN search to assign points to centers. The assignment of points to centers can be represented with a log k-bit index per data point. The cost of storing the centers grows linearly with k. 3. Orthogonal k-means with 2m centers With a compositional model one can represent cluster centers more efficiently. One such approach is to re- construct each input with an additive combination of the columns of C. To this end, instead of the 1-of-k encoding in (2), we let b be a general m-bit vector, b ∈ Hm ≡ {0, 1}m, (a2n)d, wCe e∈ l Rt bp× bme. aA gse nsuerchal, meac-bhi tc vluesctteorr c, ebnt ∈er H His th≡e sum o}f a asundbs Cet o∈f Rthe columns of C. There are 2m possible subsets, and therefore k = 2m centers. While the number of parameters in the quantizer is linear in m, the number of centers increases exponentially. While efficient in representing cluster centers, the approach is problematic, because solving bˆ = abrg∈Hmmin?x − Cb?22,(3) is intractable; i.e., the discrete optimization is not submodular. Obviously, for small 2m one could generate all possible centers and then perform NN search to find the optimal solution, but this would not scale well to large values of m. One key observation is that if the columns of C are orthogonal, then optimization (3) becomes tractable. To explain this, without loss of generality, assume the bits belong to {−1, 1} instead of {0, 1}, i.e., b? ∈ Hm? ≡ {−1, 1}m. tToh e{−n, ?x − Cb??22 = xTx + b?TCTCb? − 2xTCb? .(4) For diagonal CTC, the middle quadratic term on the RHS becomes trace(CTC), independent of b?. As a consequence, when C has orthogonal columns, abr?g∈mH?min?x − Cb??22 = sgn(CTx) ,(5) where sgn(·) is the element-wise sign function. Cluster centers are depicted by dots, and cluster boundaries by dashed lines. (Left) Clusters formed by a 2-cube with no rotation, scaling or translation; centers = {b? |b? ∈ H2? }. (Right) Rotation, scaling and translation are used to reduce distances between data points and cluster centers; centers = {μ + RDb? | b? ∈ H2? }. To reduce quantization error further we can also introduce an offset, denoted μ, to translate x. Taken together with (5), this leads to the loss function for the model we call orthogonal k-means (ok-means):2 ?ok-means(C,μ) = x?∈Db?m∈iHn?m?x − μ − Cb??22. (6) Clearly, with a change of variables, b? = 2b −1, we can define new versions of μ and C, with ide=ntic 2abl −lo1s,s, w wfoer c wanh dicehthe unknowns are binary, but in {0, 1}m. T uhnek nook-wmnesa anrse quantizer uetn icnod {e0,s 1ea}ch data point as a vertex of a transformed m-dimensional unit hypercube. The transform, via C and μ, maps the hypercube vertices onto the input feature space, ideally as close as possible to the data points. The matrix C has orthogonal columns and can therefore be expressed in terms of rotation and scaling; i.e., C ≡ RD, where R ∈ Rp×m has orthonormal cinoglu;m i.ne.s, ( CRT ≡R = R DIm,) w, ahnerde eD R Ris ∈ diagonal and positive definite. The goal of learning is to find the parameters, R, D, and μ, which minimize quantization error. Fig. 1 depicts a small set of 2D data points (red x’s) and two possible quantizations. Fig. 1 (left) depicts the vertices of a 2-cube with C = I2 and zero translation. The cluster regions are simply the four quadrants of the 2D space. The distances between data points and cluster centers, i.e., the quantization errors, are relatively large. By comparison, Fig. 1 (right) shows how a transformed 2-cube, the full model, can greatly reduce quantization errors. 3.1. Learning ok-means To derive the learning algorithm for ok-means we first rewrite the objective in matrix terms. Given n data points, let X = [x1, x2, · · · , xn] ∈ Rp×n. Let B? ∈ {−1, 1}m×n denote the corresponding ∈clu Rster assignment {co−ef1f,ic1i}ents. Our 2While closely related to ITQ, we use the the relationship to k-means. term ok-means to emphasize 333000111 686 goal is to find the assignment coefficients B? and the transformation parameters, namely, the rotation R, scaling D, and translation μ, that minimize ?ok-means(B?, R, D,μ) = ?X − μ1T − RDB??f2 (7) = ?X? − RDB??f2 (8) where ?·?f denotes the Frobenius norm, X? ≡ X − μ1T, R iws hceornes ?tr·a?ined to have orthonormal columns≡ ≡(R XTR − μ=1 Im), and D is a positive diagonal matrix. Like k-means, coordinate descent is effective for optimizing (8). We first initialize μ and R, and then iteratively optimize ?ok-means with respect to B?, D, R, and μ: • Optimize B? and D: With straightforward algebraic manipulation, one can show that ?ok-means = ?RTX?−DB??f2 + ?R⊥TX??f2 , (9) where columns of R⊥ span the orthogonal complement of the column-space of R (i.e., the block matrix [R R⊥] is orthogonal). It follows that, given X? and R, we can optimize the first term in (9) to solve for B? and D. Here, DB? is the leastsquares approximation to RTX?, where RTX? and DB? × are m n. Further, the ith row of DB? can only contain earleem men×tsn .fr Fomur t{h−erd,i t,h +e dii} where di = Dii. Wherever tehleem corresponding delement} }o fw RheTrXe? d is positive (negative) we should put a positive (negative) value in DB?. The optimal di is determined by the mean absolute value of the elements in the ith row of RTX?: • • B? ← sgn ?RTX?? (10) D ← Diag?(mroewan?(abs(RTX?))) (11) Optimize R: Using the original objective (8), find R that minimizes ?X? −RA?f2 , subject to RTR = Im, and Am n≡i mDizBes?. ?TXhis i−s equivalent to an Orthogonal Procrustes problem [15], and can be solved exactly using SVD. In particular, by adding p − m rows of zeros to the bottom poaf rDtic, uDlaBr, bb eyc aodmdeins p p× − n. mT rhoewn sR o ifs z square a tnhde orthogoonf aDl ,a DndB can boem eessti pm ×at end. Twhiethn RSV isD s. qBuuatr es ainncde oDrtBho gisdegenerate we are only interested in the first m columns of R; the remaining columns are unique only up to rotation of the null-space.) Optimize μ: Given R, B? and D, the optimal μ is given by the column average of X −RDB?: μ ← mcoeluamn(X−RDB?) (12) 3.2. Approximate nearest neighbor search One application of scalable quantization is ANN search. Before introducing more advanced quantization techniques, we describe some experimental results with ok-means on Euclidean ANN search. The benefits of ok-means for ANN search are two-fold. Storage costs for the centers is reduced to O(log k), from O(k) with k-means. Second, substantial speedups are possible by exploiting fast methods for NN search on binary codes in Hamming space (e.g., [13]). Generally, in terms of a generic quantizer q(·), there are twoG neanteurraalll ways etrom messti omfa ate g ethneer dicis qtaunanceti z beetrw qe(·e)n, ttwheor vectors, v and u [7]. Using the Symmetric quantizer distance (SQD) ?v−u? is approximated by ?q(v)−q(u) ? . Using the Asymmetric quantizer doixsimtanacteed (A byQ ?Dq()v, only one o Uf sthineg tw thoe vectors is quantized, and ?v−u? is estimated as ?v−q(u) ? . vWechtiloer sS iQs qDu might b, aen slightly f?as itse ers t iom compute, vA−QqD(u i)?n-. curs lower quantization errors, and hence is more accurate. For ANN search, in a pre-processing stage, each database entry, u, is encoded into a binary vector corresponding to the cluster center index to which u is assigned. At test time, the queries may or may not be encoded into indices, depending on whether one uses SQD or AQD. In the ok-means model, the quantization of an input x is straightforwardly shown to be qok(x) = μ + RD sgn(RT(x − μ)) . (13) The corresponding m-bit cluster index is sgn(RT(x − μ)). Given two indices, namely b?1 , b?2 ∈ {−1, +1}m,( txhe − symmetric ok-means quantizer distance∈ ∈is { SQDok(b?1, b?2) = ?μ + RDb?1 − μ − RDb?2 ?22 = ?D(b?1 − b?2)?22 .(14) In effect, SQDok is a weighted Hamming distance. It is the sum of the squared diagonal entries of D corresponding to bits where b?1 and b?2 differ. Interestingly, in our experiments with ok-means, Hamming and weighted Hamming distances yield similar results. Thus, in ok-means experiments we simply report results for Hamming distance, to facilitate comparisons with other techniques. When the scaling in ok-means is constrained to be isotropic (i.e., D = αIm for α ∈ R+), then SQDok becomes a constant multiple off othre α αu ∈sua Rl Hamming distance. As discussed in Sec. 5, this isotropic ok-means is closely related to ITQ [5]. The ok-means AQD between a feature vector x and a cluster index b?, is defined as AQDok(x, b?) = ?x −μ − RDb??22 = ?RTx? − Db??22 + ?R⊥Tx??22 , (15) where x? = x−μ. For ANN search, in comparing distances from query x t−o a .d Faotars AetN oNf binary i,n idni ccoesm, ptharei snegc odnisdta tnecrems on the RHS of (15) is irrelevant, since it does not depend on b?. Without this term, AQDok becomes a form of asymmetric Hamming (AH) distance between RTx? and b?. While previous work [6] discussed various ad hoc AH distance measures for binary hashing, interestingly, the optimal AH distance for ok-means is derived directly in our model. 333000111977 1M SIFT, 64−bit encoding (k = 264) lRc@aeR0 . 206148 10 RIoPT1Qk K−Q m( AHeDa)Hn )s (AH)10K Figure 2. Euclidean ANN retrieval results for different methods and distance functions on the 1M SIFT dataset. 3.3. Experiments with ok-means Following [7], we report ANN search results on 1M SIFT, a corpus of 128D SIFT descriptors with disjoint sets of 100K training, 1M base, and 10K test vectors. The training set is used to train models. The base set is the database, and the test points are queries. The number of bits m is typically less than p, but no pre-processing is needed for dimensionality reduction. Rather, to initialize learning, R is a random rotation of the first m principal directions of the training data, and μ is the mean of the data. For each query we find R nearest neighbors, and compute Recall@R, the fraction of queries for which the ground-truth Euclidean NN is found in the R retrieved items. Fig. 2 shows the Recall@R plots for ok-means with Hamming (H) ≈ SQDok and asymmetric Hamming (AH) H≡a mAmQiDngok ( Hd)is t≈an SceQ, vs. ITQ [5] and PQ [7]. The PQ ≡met AhoQdD exploits a more complex asymmetric distance func- tion denoted AD ≡ AQDck (defined in Sec. 6. 1). Note first tthioant od ke-nmoeteadns A improves upon ITQ, with both Hamming and asymmetric Hamming distances. This is due to the scaling parameters (i.e., D) in ok-means. If one is interested in Hamming distance efficient retrieval, ok-means is prefered over ITQ. But better results are obtained with the asymmetric distance function. Fig. 2 also shows that PQ achieves superior recall rates. This stems from its joint encoding of multiple feature dimensions. In ok-means, each bit represents a partition ofthe feature space into two clusters, separated by a hyperplane. The intersection of m orthogonal hyperplanes yields 2m regions. Hence we obtain just two clusters per dimension, and each dimension is encoded independently. In PQ, by comparison, multiple dimensions are encoded jointly, with arbitrary numbers of clusters. PQ thereby captures statistical dependencies among different dimensions. We next extend ok-means to jointly encode multiple dimensions. 4. Cartesian k-means In the Cartesian k-means (ck-means) model, each region center is expressed parametrically as an additive combination of multiple subcenters. Let there be m sets of subcen- Fidg2ure31.Ddep4icton5fCartde?1si2nqdu?4ati5z?3aton 4qDck(da)t= ,?wd i?5134t?h the first (last) two dimensions sub-quantized on the left (right). Cartesian k-means quantizer denoted qck, combines the subquantizations in subspaces, and produces a 4D reconstruction. ters, each with h elements.3 Let C(i) be a matrix whose columns comprise the elements of the ith subcenter set; C(i) ∈ Rp×h. Finally, assume that each cluster center, c, is the∈ sum of exactly one element from each subcenter set: = ?C(i)b(i) m c i?= ?1 , (16) where b(i) ∈ H1/h is a 1-of-h encoding. As a conc∈re Hte example (see Fig. 3), suppose we are given 4D inputs, x ∈ R4, and we split each datum into m = 2 parts: z(1) = ?I2 0? x , and Then, suppose w?e z(2) = ?0 I2? x .(17) qu?antize each part, z(?1) and? z(2) , sepa- × rately. As depicted in Fig. 3 (left and middle), we could use h = 5 subcenters for each one. Placing the corresponding subcenters in the columns of 4 5 matrices C(1) and C(2) , C(1)=?d1d20d2×35d4d5?, C(2)=?d?1d?20d2×?35d?4d?5?, we obtain a model (16) that provides 52 possible centers with which to quantize the data. More generally, the total number of model centers is k = hm. Each center is a member of the Cartesian product of the subcenter sets, hence the name Cartesian k-means. Importantly, while the number of centers is hm, the number of subcenters is only mh. The model provides a super-linear number of centers with a linear number of parameters. The learning objective for Cartesian k-means is ?ck-means(C) =x?∈D{b(mi)}inim=1???x −i?=m1C(i)b(i)??22 where b(i) ∈ H1/h, and C ≡ [C(1), ··· , (18) C(m)] ∈ Rp×mh. [b(1)T, ··· ,b(m)T] If we let bT ≡ then the second sum in (18) can be expressed succinctly as Cb. 3While here we assume a fixed cardinality for all subcenter sets, the model is easily extended to allow sets with different cardinalities. 333000112088 The key problem with this formulation is that the min- imization of (18) with respect to the b(i) ’s is intractable. Nevertheless, motivated by orthogonal k-means (Sec. 3), encoding can be shown to be both efficient and exact if we impose orthogonality constraints on the sets of subcenters. To that end, assume that all subcenters in different sets are pairwise orthogonal; i.e., ∀i,j | i = j C(i)TC(j) = 0h×h .(19) Each subcenter matrix C(i) spans a linear subspace of Rp, and the linear subspaces for different subcenter sets do not intersect. Hence, (19) implies that only the subcenters in C(i) can explain the projection of x onto the C(i) subspace. In the example depicted in Fig. 3, the input features are simply partitioned (17), and the subspaces clearly satisfy the orthogonality constraints. It is also clear that C ≡ [ C(1) C(2)] is block diagonal, Iwtit ihs 2 a ×lso o5 c bleloacrks t,h adte Cnote ≡d D(1) and D(]2 i)s . bTlohec quantization error t×he5re bfolorcek bse,c doemnoeste ?x − Cb?22 = ???zz((12))?−?D0(1) D0(2)? ?b ( 21) ? ?2 = ?????z(1)−D(1)b(1)??2+???z(2)−D(2??)b(2)??2. In words, the squa??zred quantization?? erro??r zis the sum of t??he squared errors on the subspaces. One can therefore solve for the binary coefficients of the subcenter sets independently. In the general case, assuming (19) is satisfied, C can be expressed as a product R D, where R has orthonormal columns, and D is block diagonal; i.e., C = R D where Ra=nd[hRe(n1c),e·C(,i)R=(mR)]i,Dan(di).DW=i⎢t⎡⎣⎢hDs0i.(1≡)Dra0(n2)k. C.(Di)0(.m,i)t⎦⎥ ⎤fo,l(2w0)s that D(i) ∈ Rsi×h and R(i) ∈ Rp×≡sira. Clearly, ? si ≤ p, because of∈ ∈th Re orthogonality ∈con Rstraints. Replacing C(i) with R(i)D(i) in the RHS of (18?), we find ?x−Cb?22 = ?m?z(i)−D(i)b(i)?22+?R⊥Tx?22, (21) i?= ?1 where z(i)≡R(i)Tx, and R⊥is the orthogonal complement of R. This≡ ≡shRows that, with orthogonality constraints (19), the ck-means encoding problem can be split into m independent sub-encoding problems, one for each subcenter set. To find the b(i) that minimizes ??z(i) we perform NN search with z(i) again??st h si-dim vec??tors in D(i) . This entails a cost of O(hsi).? Fortunately, all? the elements of b can be found very efficiently, in O(hs), where s ≡ ? si. If we also include the cost of rotating x to −D(i)b(i)?22, Taocbkml-emt1he.aoAnds um#ceh2nkmrtyeofskm-#lobgeiatknsh,cOo-rm( pOec(2oamkn+spt),hpan)dkO- m(pOecosa(n+mst(hipns)khtesro)m of number of centers, number of bits needed for indices (i.e., log #centers), and the storage cost of representation, which is the same as the encoding cost to convert inputs to indices. The last column shows the costs given an assumption that C has a rank of s ≥ m. obtain each z(i) , the total encoding cost is O(ps + hs), i.e., O(p2+hp). Alternatively, one could perform NN search in p-dim C(i) ’s, to find the b(i) ’s, which costs O(mhp). Table 1 summerizes the models in terms of their number of centers, index size, and cost of storage and encoding. 4.1. Learning ck-means We can re-write the ck-means objective (18) in matrix form with the Frobenius norm; i.e., ?ck-means(R, D, B) = ? X − RDB ?f2 (22) where the columns of X and B comprise the data points and the subcenter assignment coefficients. The input to the learning algorithm is the training data X, the number of subcenter sets m, the cardinality of the subcenter sets h, and an upper bound on the rank of C, i.e., s. In practice, we also let each rotation matrix R(i) have the same number of columns, i.e., si = s/m. The outputs are the matrices {R(i) } and {D(i) } that provide a local minima of (22). Learning begins hwaitth p trohev idneit aia lloizcaatlio mni noimf Ra oanf d(2 D2)., followed by iterative coordinate descent in B, D, and R: • Optimize B and D: With R fixed, the objective is given by (21) where ?R⊥TX?f2 R(i)TX, is constant. Given data pro- jections Z(i) ≡ to optimize for B and D we perform one step oRf k-means for each subcenter set: – Assignment: Perform NN searches for each subcenter set to find the assignment coefficients, B(i) , B(i)← arBg(mi)in?Z(i)− D(i)B(i)?f2 – • Update: D(i)← arDg(mi)in?Z(i)− D(i)B(i)?f2 Optimize R: Placing the D(i) ’s along the diagonal of D, as in (20), and concatenating B(i) ’s as rows of B, [B(1)T, ...,B(m)T], i.e., BT = the optimization of R reduces to the orthogonal Procrustes problem: R ← argRmin?X − RDB?f2. In experiments below, R ∈ Rp×p, and rank(C) ≤ p is unIcnon esxtpraeirniemde. tFso rb high-dimensional adnadta r awnhke(rCe ) ra ≤nk( pX is) ?np, fnostrr efficiency ri th may dbime eusnesfiuoln atol dcoantast wrahiner er anr akn(kC(X). 333000112199 5. Relations and related work As mentioned above, there are close mathematical relationships between ok-means, ck-means, ITQ for binary hashing [5], and PQ for vector quantization [7]. It is instructive to specify these relationships in some detail. Iterative Quantization vs. Orthogonal k-means. ITQ [5] is a variant of locality-sensitive hashing, mapping data to binary codes for fast retrieval. To extract m-bit codes, ITQ first zero-centers the data matrix to obtain X?. PCA is then used for dimensionality reduction, fromp down to m dimensions, after which the subspace representation is randomly rotated. The composition of PCA and the random rotation can be expressed as WTX? where W ∈ Rp×m. ITQ then solves for the m×m rotation mawthriex,r eR W, t ∈hatR minimizes ?ITQ(B?, R) = ?RTWTX?− B??f2 , s.t. RTR = Im×m, (23) where B? ∈ {−1, +1}n×p. ITQ ro∈tate {s− t1h,e+ subspace representation of the data to match the binary codes, and then minimizes quantization error within the subspace. By contrast, ok-means maps the binary codes into the original input space, and then considers both the quantization error within the subspace and the out-of-subspace projection error. A key difference is the inclusion of ?R⊥X? ?f2 in the ok-means objective (9). This is important s ?inRce one can often reduce quantization errors by projecting out significant portions of the feature space. Another key difference between ITQ and ok-means is the inclusion of non-uniform scaling in ok-means. This is important when the data are not isotropic, and contributes to the marked improvement in recall rates in Fig. 2. Orthogonal k-means vs. Cartesian k-means. We next show that ok-means is a special case of ck-means with h = 2, where each subcenter set has two elements. To this end, let C(i) = and let b(i) = be the ith subcenter matrix selection vector. Since b(i) is a 1-of-2 encoding (?10? or ?01?), it follows that: [c(1i) c2(i)], b1(i)c(1i)+ b(2i)c2(i) = [b(1i) b2(i)]T c(1i)+2 c2(i)+ bi?c1(i)−2 c2(i), (24) b1(i) − b2(i) where bi? ≡ ∈ {−1, +1}. With the following setting of t≡he bok-m−e abns parameters, μ=?i=m1c1(i)+2c(2i),andC=?c1(1)−2c2(1),...,c1(m)−2c2(m)?, it should be clear that ?i C(i)b(i) = μ + Cb?, where b? ∈ {−1, +1}m, and b??i is the ith bit of b?, used in (24). Similarly, one can also map ok-means parameters onto corresponding subcenters for ck-means. Thus, there is a 1-to-1 mapping between the parameterizations of cluster centers in ok-means and ck-means for h = 2. The benefits of ok-means are its small number of parameters, and its intuitive formulation. The benefit of the ck-means generalization is its joint encoding of multiple dimensions with an arbitrary number of centers, allowing it to capture intrinsic dependence among data dimensions. Product Quantization vs. Cartesian k-means. PQ first splits the input vectors into m disjoint sub-vectors, and then quantizes each sub-vector separately using a subquantizer. Thus PQ is a special case of ck-means where the rotation R is not optimized; rather, R is fixed in both learning and retrieval. This is important because the sub-quantizers act independently, thereby ignoring intrasubspace statistical dependence. Thus the selection of subspaces is critical for PQ [7, 8]. J e´gou et al. [8] suggest using PCA, followed by random rotation, before applying PQ to high-dimensional vectors. Finding the rotation to minimize quantization error is mentioned in [8], but it was considered too difficult to estimate. Here we show that one can find a rotation to minimize the quantization error. The ck-means learning algorithm optimizes sub-quantizers in the inner loop, but they interact in an outer loop that optimizes the rotation (Sec. 4.1). 6. Experiments 6.1. Euclidean ANN search Euclidean ANN is a useful task for comparing quantization techniques. Given a database of ck-means indices, and a query, we use Asymmetric and Symmetric ck-means quantizer distance, denoted AQDck and SQDck. The AQDck between a query x and a binary index b ≡ ?b(1)T, ...,b(m)T?T, derived in (21), is AQDck(x,b) ?m?z(i)−D(i)b(i)?22+ ?R⊥Tx?22.(25) = i?= ?1 ?z(i)−D(i)b(i)?22is the distance between the ithpro- Here, jection?? of x, i.e., z(i) , ?a?nd a subcenter projection from D(i) selecte?d by b(i) . Give?n a query, these distances for each i, and all h possible values of b(i) can be pre-computed and stored in a query-specific h×m lookup table. Once created, tshtoer lookup qtaubelery i-ss pusecedif itoc compare akullp pda tatbablea.se O points etaot tehde, query. So computing AQDck entails m lookups and additions, somewhat more costly than Hamming distance. The last term on the RHS of (25) is irrelevant for NN search. Since PQ is a special case of ck-means with pre-defined subspaces, the same distances are used for PQ (cf. [7]). The SQDck between binary codes b1 and b2 is given by SQDck(b1,b2) = ?m?D(i)b1(i)− D(i)b2(i)?22.(26) i?= ?1 Since b1(i) and b(2i) are 1-of-?h encodings, an m×h?×h lookup table can be createadre t 1o- sft-ohre e nacllo pairwise msu×bh×-dhis ltaonockeusp. 333000222200 1M SIFT, 64−bit encoding (k = 264) 1M GIST, 64−bit encoding (k = 264) 1B SIFT, 64−bit encoding (k = 264) Re@acl0 .4186021RcIPoTQk0−Q m( SAeDaH)n s (SADH)1K0 .1642081 0RIocP1TkQ −K m (SAeDHa)n s (ASHD1)0K .186420 1 R0IoPc1TkQ −KQm( ASe DaHn) s (SADH1)0K Figure 4. Euclidean NN recall@R (number of items retrieved) based on different quantizers and corresponding distance functions on the 1M SIFT, 1M GIST, and 1B SIFT datasets. The dashed curves use symmetric distance. (AH ≡ AQDok, SD ≡ SQDck, AD ≡ AQDck) While the cost of computing SQDck is the same as AQDck, SQDck could also be used to estimate the distance between the indexed database entries, for diversifying the retrieval results, or to detect near duplicates, for example. Datasets. We use the 1M SIFT dataset, as in Sec. 3.3, along with two others, namely, 1M GIST (960D features) and 1B SIFT, both comprising disjoint sets of training, base and test vectors. 1M GIST has 500K training, 1M base, and 1K query vectors. 1B SIFT has 100M training, 1B base, and 10K query points. In each case, recall rates are averaged over queries in test set for a database populated from the base set. For expedience, we use only the first 1M training points for the 1B SIFT experiments. Parameters. In ANN experiments below, for both ckmeans and PQ, we use m = 8 and h = 256. Hence the number of clusters is k = 2568 = 264, so 64-bits are used as database indices. Using h = 256 is particularly attractive because the resulting lookup tables are small, encoding is fast, and each subcenter index fits into one byte. As h increases we expect retrieval results to improve, but encoding and indexing of a query to become slower. Initialization. To initialize the Di’s for learning, as in kmeans, we simply begin with random samples from the set of Zi’s (see Sec. 4.1). To initialize R we consider the different methods that J e´gou et al. [7] proposed for splitting the feature dimensions into m sub-vectors for PQ: (1) natural: sub-vectors comprise consecutive dimensions, (2) structured: dimensions with the same index modulo 8 are grouped, and (3) random: random permutations are used. For PQ in the experiments below, we use the orderings that produced the best results in [7], namely, the structured ordering for 960D GIST, and the natural ordering for 128D SIFT. For learning ck-means, R is initialized to the identity with SIFT corpora. For 1M GIST, where the PQ ordering is significant, we consider all three orderings to initialize R. Results. Fig. 4 shows Recall@R plots for ck-means and PQ [7] with symmetric and asymmetric distances (SD ≡ PSQQD [7c]k wanitdh A syDm m≡e rAicQ aDncdk) a on tmhee t3r cda dtaissteatns.c Tsh (eS Dhor ≡izontal axainsd represents tQheD number of retrieved items, R, on a log-scale. The results consistently favor ck-means. On 1M GIST, 64−bit encoding (k = 264) lae@Rc0 .261084 1R0Pc Qk −1m(K 321e )a (A nD s )(132) (A D )10K Figure 5. PQ and ck-means results using natural (1), structured (2), and random (3) ordering to define the (initial) subspaces. the high-dimensional GIST data, ck-means with AD significantly outperforms other methods; even ck-means with SD performs on par with PQ with AD. On 1M SIFT, the Recall@ 10 numbers for PQ and ck-means, both using AD, × are 59.9% and 63.7%. On 1B SIFT, Recall@ 100 numbers are 56.5% and 64.9%. As expected, with increasing dataset size, the difference between methods is more significant. In 1B SIFT, each feature vector is 128 bytes, hence a total of 119 GB. Using any method in Fig. 4 (including ckmeans) to index the database into 64 bits, this storage cost reduces to only 7.5 GB. This allows one to work with much larger datasets. In the experiments we use linear scan to find the nearest items according to quantizer distances. For NN search using 10K SIFT queries on 1B SIFT this takes about 8 hours for AD and AH and 4 hours for Hamming distance on a 2 4-core computer. Search can be sped up significantly; using a coarse tienri.tia Sl quantization sapnedd an pin svigenritefidfile structure for AD and AH, as suggested by [7, 1], and using the multi-index hashing method of [13] for Hamming distance. In this paper we did not implement these efficiencies as we focus primarily on the quality of quantization. Fig. 5 compares ck-means to PQ when R in ck-means is initialized using the 3 orderings of [7]. It shows that ckmeans is superior in all cases. Simiarly interesting, it also shows that despite the non-convexity ofthe optimization objective, ck-means learning tends to find similarly good encodings under different initial conditions. Finally, Fig. 6 compares the methods under different code dimensionality. 333000222311 1M GIST, encoding with 64, 96, and 128 bits Rl@ace0 .814206 1 R0cP kQ − m1 69e4216a−8 Knb −itsb 1t69248− bit10K Figure 6. PQ and ck-means results using different number of bits for encoding. In all cases asymmetric distance is used. Table2.Rcokg-nmiteoamPnQCesac(ondksue=r(bkaoc416y=0ko26)n40t2h[eC]IFAR7c598-u1.72096r%a tcesyuingdfferent codebook learning algorithms. 6.2. Learning visual codebooks While the ANN seach tasks above require too many clusters for k-means, it is interesing to compare k-means and ck-means on a task with a moderate number of clusters. To this end we consider codebook learning for bag-ofword√s models [3, 10]. We use ck-means with m = 2 and h = √k, and hence k centers. The main advantage of ck- × here is that finding the closest cluster center is done in O(√k) time, much faster than standard NN search with k-means in O(k). Alternatives for k-means, to improve efficiency, include approximate k-means [14], and hierarchical k-means [12]. Here we only compare to exact k-means. CIFAR-10 [9] comprises 50K training and 10K test images (32 32 pixels). Each image is one of 10 classes (airplane, 3b2i×rd,3 car, cat, )d.e Eera,c dog, frog, sh oonrsee o, ship, laansds etrsu (caki)r.We use publicly available code from Coates et al [2], with changes to the codebook learning and cluster assignment modules. Codebooks are built on 6×6 whitened color image patches. .O Cnoed histogram i sb ucirleta otend 6 per image quadrant, aagnde a linear SVM is applied to 4k-dim feature vectors. Recognition accuracy rates on the test set for different models and k are given in Table 2. Despite having fewer parameters, ck-means performs on par or better than kmeans. This is consistent for different initializations of the algorithms. Although k-means has higher fedility than ckmeans, with fewer parameters, ck-means may be less susceptible to overfitting. Table 2, also compares with the approach of [19], where PQ without learning a rotation is used for clustering features. As expected, learning the rotation has a significant impact on recognition rates, outperforming all three initializations of PQ. mean√s 7. Conclusions We present the Cartesian k-means algorithm, a generalization of k-means with a parameterization of the clus- ter centers such that number of centers is super-linear in the number of parameters. The method is also shown to be a generalization of the ITQ algorithm and Product Quantization. In experiments on large-scale retrieval and codebook learning for recognition the results are impressive, outperforming product quantization by a significant margin. An implementation of the method is available at https : / / github . com/norou z i ckmeans . / References [1] A. Babenko and V. Lempitsky. The inverted multi-index. CVPR, 2012. [2] A. Coates, H. Lee, and A. Ng. An analysis of single-layer networks in unsupervised feature learning. AISTATS, 2011. [3] G. Csurka, C. Dance, L. Fan, J. Willamowski, and C. Bray. Visual categorization with bags of keypoints. ECCV Workshop Statistical Learning in Computer Vision, 2004. [4] T. Ge, K. He, Q. Ke, and J. Sun. Optimized product quantization for approximate nearest neighbor search. CVPR, 2013. [5] Y. Gong and S. Lazebnik. Iterative quantization: A procrustean approach to learning binary codes. CVPR, 2011. [6] A. Gordo and F. Perronnin. Asymmetric distances for binary embeddings. CVPR, 2011. [7] H. J ´egou, M. Douze, and C. Schmid. Product quantization for nearest neighbor search. IEEE Trans. PAMI, 2011. [8] H. J ´egou, M. Douze, C. Schmid, and P. P ´erez. Aggregating local descriptors into a compact image representation. CVPR, 2010. [9] A. Krizhevsky. Learning multiple layers of features from tiny images. MSc Thesis, Univ. Toronto, 2009. [10] S. Lazebnik, C. Schmid, and J. Ponce. Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories. CVPR, 2006. [11] S. P. Lloyd. Least squares quantization in pcm. IEEE Trans. IT, 28(2): 129–137, 1982. [12] D. Nister and H. Stewenius. Scalable recognition with a vocabulary tree. CVPR, 2006. [13] M. Norouzi, A. Punjani, and D. J. Fleet. Fast search in hamming space with multi-index hashing. CVPR, 2012. [14] J. Philbin, O. Chum, M. Isard, J. Sivic, and A. Zisserman. Object retrieval with large vocabularies and fast spatial matching. CVPR, 2007. [15] P. Sch o¨nemann. A generalized solution of the orthogonal procrustes problem. Psychometrika, 3 1, 1966. [16] J. Sivic and A. Zisserman. Video Google: A text retrieval approach to object matching in videos. ICCV, 2003. [17] J. Tenenbaum and W. Freeman. Separating style and content with bilinear models. Neural Comp., 12: 1247–1283, 2000. [18] A. Torralba, K. Murphy, and W. Freeman. Sharing visual features for multiclass and multiview object detection. IEEE T. PAMI, 29(5):854–869, 2007. [19] S. Wei, X. Wu, and D. Xu. Partitioned k-means clustering for fast construction of unbiased visual vocabulary. The Era of Interactive Media, 2012. 333000222422
same-paper 4 0.89228749 255 cvpr-2013-Learning Separable Filters
Author: Roberto Rigamonti, Amos Sironi, Vincent Lepetit, Pascal Fua
Abstract: Learning filters to produce sparse image representations in terms of overcomplete dictionaries has emerged as a powerful way to create image features for many different purposes. Unfortunately, these filters are usually both numerous and non-separable, making their use computationally expensive. In this paper, we show that such filters can be computed as linear combinations of a smaller number of separable ones, thus greatly reducing the computational complexity at no cost in terms of performance. This makes filter learning approaches practical even for large images or 3D volumes, and we show that we significantly outperform state-of-theart methods on the linear structure extraction task, in terms ofboth accuracy and speed. Moreover, our approach is general and can be used on generic filter banks to reduce the complexity of the convolutions.
5 0.89126825 421 cvpr-2013-Supervised Kernel Descriptors for Visual Recognition
Author: Peng Wang, Jingdong Wang, Gang Zeng, Weiwei Xu, Hongbin Zha, Shipeng Li
Abstract: In visual recognition tasks, the design of low level image feature representation is fundamental. The advent of local patch features from pixel attributes such as SIFT and LBP, has precipitated dramatic progresses. Recently, a kernel view of these features, called kernel descriptors (KDES) [1], generalizes the feature design in an unsupervised fashion and yields impressive results. In this paper, we present a supervised framework to embed the image level label information into the design of patch level kernel descriptors, which we call supervised kernel descriptors (SKDES). Specifically, we adopt the broadly applied bag-of-words (BOW) image classification pipeline and a large margin criterion to learn the lowlevel patch representation, which makes the patch features much more compact and achieve better discriminative ability than KDES. With this method, we achieve competitive results over several public datasets comparing with stateof-the-art methods.
6 0.86016738 362 cvpr-2013-Robust Monocular Epipolar Flow Estimation
7 0.85528064 70 cvpr-2013-Bottom-Up Segmentation for Top-Down Detection
8 0.85517228 446 cvpr-2013-Understanding Indoor Scenes Using 3D Geometric Phrases
10 0.85305667 445 cvpr-2013-Understanding Bayesian Rooms Using Composite 3D Object Models
11 0.85262758 104 cvpr-2013-Deep Convolutional Network Cascade for Facial Point Detection
12 0.85235637 61 cvpr-2013-Beyond Point Clouds: Scene Understanding by Reasoning Geometry and Physics
13 0.85118377 424 cvpr-2013-Templateless Quasi-rigid Shape Modeling with Implicit Loop-Closure
14 0.8511827 372 cvpr-2013-SLAM++: Simultaneous Localisation and Mapping at the Level of Objects
15 0.85093188 256 cvpr-2013-Learning Structured Hough Voting for Joint Object Detection and Occlusion Reasoning
16 0.85089713 69 cvpr-2013-Boosting Binary Keypoint Descriptors
17 0.85089475 425 cvpr-2013-Tensor-Based High-Order Semantic Relation Transfer for Semantic Scene Segmentation
18 0.85082215 67 cvpr-2013-Blocks That Shout: Distinctive Parts for Scene Classification
19 0.85068709 132 cvpr-2013-Discriminative Re-ranking of Diverse Segmentations
20 0.85037422 350 cvpr-2013-Reconstructing Loopy Curvilinear Structures Using Integer Programming