cvpr cvpr2013 cvpr2013-404 knowledge-graph by maker-knowledge-mining

404 cvpr-2013-Sparse Quantization for Patch Description

Source: pdf

Author: Xavier Boix, Michael Gygli, Gemma Roig, Luc Van_Gool

Abstract: The representation of local image patches is crucial for the good performance and efficiency of many vision tasks. Patch descriptors have been designed to generalize towards diverse variations, depending on the application, as well as the desired compromise between accuracy and efficiency. We present a novel formulation of patch description, that serves such issues well. Sparse quantization lies at its heart. This allows for efficient encodings, leading to powerful, novel binary descriptors, yet also to the generalization of existing descriptors like SIFTorBRIEF. We demonstrate the capabilities of our formulation for both keypoint matching and image classification. Our binary descriptors achieve state-of-the-art results for two keypoint matching benchmarks, namely those by Brown [6] and Mikolajczyk [18]. For image classification, we propose new descriptors that perform similar to SIFT on Caltech101 [10] and PASCAL VOC07 [9].

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 Patch descriptors have been designed to generalize towards diverse variations, depending on the application, as well as the desired compromise between accuracy and efficiency. [sent-5, score-0.127]

2 We present a novel formulation of patch description, that serves such issues well. [sent-6, score-0.306]

3 This allows for efficient encodings, leading to powerful, novel binary descriptors, yet also to the generalization of existing descriptors like SIFTorBRIEF. [sent-8, score-0.177]

4 We demonstrate the capabilities of our formulation for both keypoint matching and image classification. [sent-9, score-0.319]

5 Our binary descriptors achieve state-of-the-art results for two keypoint matching benchmarks, namely those by Brown [6] and Mikolajczyk [18]. [sent-10, score-0.332]

6 For image classification, we propose new descriptors that perform similar to SIFT on Caltech101 [10] and PASCAL VOC07 [9]. [sent-11, score-0.127]

7 The patch descriptors should typically come with a certain degree of invariance to probable image and appearance variations, while being efficient to compute. [sent-15, score-0.372]

8 Multi-view keypoint matching would require invariance under viewpoint and lighting changes. [sent-16, score-0.199]

9 There is a surfeit of such patch descriptors by now. [sent-18, score-0.328]

10 Authors can choose the most appropriate descriptor for their task, striking a balance between accuracy and efficiency. [sent-19, score-0.058]

11 The SIFT descriptor [17] is a very popular example. [sent-20, score-0.058]

12 For keypoint matching, its discriminative power has been surpassed, e. [sent-21, score-0.155]

13 by learning its pooling regions [6, 18, 21], and so has its efficiency, e. [sent-23, score-0.067]

14 SIFT may come as an integrated part thereof, but patch descriptors can also take the form of sparse coding [27], or convolutional networks [5, 13, 15]. [sent-27, score-0.372]

15 Little is understood about the common principles underlying the different patch descriptors. [sent-28, score-0.252]

16 Often descriptors appear to be disconnected from the prior art. [sent-29, score-0.127]

17 For instance, what could we say apriori about the relative performance of descriptors, even before testing them? [sent-30, score-0.053]

18 The lack of clearcut answers to such questions has led to a plethora of descriptors, designed for specific applications. [sent-31, score-0.053]

19 In this paper, we introduce a new, more principled formulation to patch description. [sent-32, score-0.306]

20 To emphasize its generality, we show that it can instantiate diverse descriptors, e. [sent-33, score-0.042]

21 We also take advantage of the capabilities of our formulation to design novel, more discriminative and computationally efficient (binary) descriptors. [sent-36, score-0.164]

22 Our formulation is based on sparse quantization (SQ) [4], which is the quantization into a set of k-sparse vectors. [sent-37, score-0.585]

23 SQ supports efficiency because it can be computed with a simple sorting and it can yield binary descriptors. [sent-38, score-0.109]

24 In a series of experiments, we report results on both, keypoint matching and image classification tasks. [sent-39, score-0.155]

25 In keypoint matching we achieve state-of-the-art results with a binary descriptor on the Brown [6] and Mikolajczyk [18] datasets. [sent-40, score-0.263]

26 It will serve as a basis for our new patch description formulation, introduced in the next section. [sent-44, score-0.301]

27 0 = k}, which is th=e s {et0 ,o1f} binary vect{o0r,s1 }with: :k ? [sent-56, score-0.05]

28 tTohres cardinality oefn |tBs qks |e ti tso equal ntod ? [sent-60, score-0.123]

29 es by defining the set of k-sparse vectors built from T =? [sent-66, score-0.084]

30 ctor v ∈ Rq into a codebook {ciT} ies a mapping nof o v ato? [sent-78, score-0.168]

31 These may bbeo any cof} t cheo previously eitn torfo kdu-scpeadr types: Rkq, Bkq, or Tkq. [sent-89, score-0.099]

32 An important advantage of SQ over a general quantization is that it can be computed much more efficiently. [sent-91, score-0.218]

33 The naive way to compute a general quantization is to evaluate the nearest neighbor of v in {ci}, which may be costly to compute fsot rn large ocro dofeb voo ikns { acn}d, high-dimensional v. [sent-92, score-0.32]

34 Tuehse olaft tthere sise an ex}t,e ancscioonrd oinfg gth teo trehseu flot l ion w[4i]n gto P trohep oTsiqkt soent. [sent-94, score-0.139]

35 2/√k, where s ∈ Tkq (all ivnepctuotr iss i nno rTmkq hlaizveed t,h ? [sent-101, score-0.042]

36 be the quantization into Tkq of v ∈ Rq, which is vˆ? [sent-109, score-0.218]

37 (1) |v| is the element-wise absolute value of v, and |kv-|Highest( |v|) is the set of dimension indices that indicate kw-hHicihg are t|hve| )k i sel tehmee snetts o of tihmee vnescitoonr i |nvd|i cwesith th tahte i highest vwahliucehs. [sent-121, score-0.174]

38 1 shows that the SQ into Tkq can be done with a sorting ofthe absolute value of {vi}, and preserving the sign soof vi ign othfteh output. [sent-123, score-0.177]

39 Atesv awluithe oSfQ{ vint}o, Rkq, irte hseasr minugchth leoswigenr computational cost than the nearest neighbor approach. [sent-124, score-0.049]

40 SQ for Encoding in Patch Description In this Section we introduce a new formulation for feature encoding, which is based on the principles of SQ. [sent-126, score-0.156]

41 Additionally, we show that SIFT and BRIEF descriptors are instances of our formulation. [sent-127, score-0.127]

42 General Overview of Patch Description We first review the pipeline for patch description and introduce some general terminology. [sent-130, score-0.301]

43 Our patch description framework can be decomposed into the following steps: 1) extracting local features, 2) encoding them, and 3) a spatial pooling stage. [sent-132, score-0.524]

44 Since we extract features at multiple locations, we obtain the set of features vectors {fj }. [sent-136, score-0.043]

45 Feature encoding maps e tahceh feeta toufre fe vaetucrteosr ivnectot a more s. [sent-137, score-0.305]

46 i Finally, tenheo pooling summarizes {αj } into a single vector which is the final descriptor osf { tαhe patch. [sent-140, score-0.125]

47 O siunrg fleea tvuercet encoding ifsra tmheew fionrakl mainly focuses on the mapping of fj to αj . [sent-141, score-0.319]

48 Encoding with SQ We apply exactly the same encoding for all fj, independently of the location of the patch from which they were extracted. [sent-145, score-0.357]

49 Our formulation is based on assignment-based encoding, inspired by the mid-level features of object recognition algorithms. [sent-147, score-0.105]

50 Assignment-based encodings such as Hard and Soft Assignment select the k entries of a codebook that are closest to f. [sent-148, score-0.294]

51 Then, the output vector, α, is built taking ones (Hard Assignment) or similarity measures (Soft Assignment) as the elements of α that correspond to the k selected entries of the codebook. [sent-149, score-0.101]

52 In our previous work [4], we have shown that Hard and Soft-Assignment encodings can be considered the SQ of a mapping of the input vector. [sent-151, score-0.226]

53 We adopt our formulation, which we used for mid-level features, as a starting point for our patch descriptor encoding. [sent-152, score-0.259]

54 We use our formulation, rather than the original formulation of Soft-Assignment, because it allows for the computational advantages of applying Prop. [sent-153, score-0.105]

55 The feature encoding is based on the mapping Ψ(f, {bi}), in which {bi ∈ Rq} is a set composed of Q vectors, }th),at i nliv we iinc hth {eb same space as etht ec input efdea toufre Qs f ∈ Rq. [sent-155, score-0.416]

56 Ψ maps (f, {bi}) to a vector in RQ, which contains tfhe ∈ similarity measures }be)t twoe aen ve tchteo rfe inatu Rre vector f and the vectors in the set {bi}. [sent-156, score-0.235]

57 sequel, is the gaussian kernel: K(f, bi) = Finally, the encoding of f is defined) as a SxpQ( o−f Ψ: α? [sent-164, score-0.156]

58 The equivalence to SoftAssignment can be seen from the fact that Ψ computes the similarities to the set {bi} (the so called codebook in the Soft-Assignment elit seerta {tubre} }[2 (t4h]e), saond c athlleend, invoking Prop. [sent-168, score-0.175]

59 1e for RkQ, the SQ of Ψ selects the k-highest entries in Ψ, and indicates this selection in α. [sent-169, score-0.06]

60 222888444311 We further develop this formulation by specifying the form of {bi}. [sent-172, score-0.105]

61 This will allow us to propose novel, powerful faonrdm mef ofifc {iebnt} encodings, a blluotw wal usos ttoo recover tnhoev wel,el plo-wkneorwfunl BRIEF and SIFT descriptors. [sent-173, score-0.151]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('sq', 0.371), ('tkq', 0.356), ('rq', 0.26), ('rkq', 0.237), ('quantization', 0.218), ('patch', 0.201), ('bi', 0.186), ('encodings', 0.168), ('encoding', 0.156), ('keypoint', 0.155), ('descriptors', 0.127), ('toufre', 0.105), ('formulation', 0.105), ('fj', 0.105), ('description', 0.1), ('oefn', 0.081), ('sel', 0.079), ('vi', 0.077), ('brief', 0.075), ('sift', 0.071), ('pooling', 0.067), ('codebook', 0.066), ('equivalence', 0.065), ('mikolajczyk', 0.063), ('ci', 0.062), ('entries', 0.06), ('capabilities', 0.059), ('sorting', 0.059), ('descriptor', 0.058), ('mapping', 0.058), ('proposition', 0.057), ('twoe', 0.053), ('haels', 0.053), ('nvd', 0.053), ('plethora', 0.053), ('troy', 0.053), ('iinc', 0.053), ('olaft', 0.053), ('tihmee', 0.053), ('vses', 0.053), ('ikns', 0.053), ('ttohres', 0.053), ('hceto', 0.053), ('thereof', 0.053), ('apriori', 0.053), ('idt', 0.053), ('tchteo', 0.053), ('formulati', 0.053), ('bbeo', 0.053), ('etre', 0.053), ('mef', 0.053), ('softassignment', 0.053), ('taoli', 0.053), ('arg', 0.052), ('principles', 0.051), ('binary', 0.05), ('assignment', 0.049), ('irte', 0.049), ('faonrdm', 0.049), ('cit', 0.049), ('ova', 0.049), ('wal', 0.049), ('roig', 0.049), ('thce', 0.049), ('ocro', 0.049), ('instrumental', 0.049), ('brown', 0.048), ('eme', 0.046), ('surpassed', 0.046), ('ihe', 0.046), ('cof', 0.046), ('acn', 0.046), ('tov', 0.046), ('cheo', 0.046), ('sparse', 0.044), ('invariance', 0.044), ('sise', 0.044), ('ikn', 0.044), ('tahceh', 0.044), ('tfhe', 0.044), ('invoking', 0.044), ('kq', 0.044), ('ctor', 0.044), ('etht', 0.044), ('vectors', 0.043), ('hve', 0.042), ('rfe', 0.042), ('gto', 0.042), ('wtoh', 0.042), ('instantiate', 0.042), ('vect', 0.042), ('nno', 0.042), ('ntod', 0.042), ('tahte', 0.042), ('built', 0.041), ('kof', 0.041), ('ign', 0.041), ('ose', 0.041), ('vo', 0.041), ('boix', 0.041)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.99999964 404 cvpr-2013-Sparse Quantization for Patch Description

Author: Xavier Boix, Michael Gygli, Gemma Roig, Luc Van_Gool

2 0.12889905 319 cvpr-2013-Optimized Product Quantization for Approximate Nearest Neighbor Search

Author: Tiezheng Ge, Kaiming He, Qifa Ke, Jian Sun

Abstract: Product quantization is an effective vector quantization approach to compactly encode high-dimensional vectors for fast approximate nearest neighbor (ANN) search. The essence of product quantization is to decompose the original high-dimensional space into the Cartesian product of a finite number of low-dimensional subspaces that are then quantized separately. Optimal space decomposition is important for the performance of ANN search, but still remains unaddressed. In this paper, we optimize product quantization by minimizing quantization distortions w.r.t. the space decomposition and the quantization codebooks. We present two novel methods for optimization: a nonparametric method that alternatively solves two smaller sub-problems, and a parametric method that is guaranteed to achieve the optimal solution if the input data follows some Gaussian distribution. We show by experiments that our optimized approach substantially improves the accuracy of product quantization for ANN search.

3 0.11244106 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

4 0.11018887 242 cvpr-2013-Label Propagation from ImageNet to 3D Point Clouds

Author: Yan Wang, Rongrong Ji, Shih-Fu Chang

Abstract: Recent years have witnessed a growing interest in understanding the semantics of point clouds in a wide variety of applications. However, point cloud labeling remains an open problem, due to the difficulty in acquiring sufficient 3D point labels towards training effective classifiers. In this paper, we overcome this challenge by utilizing the existing massive 2D semantic labeled datasets from decadelong community efforts, such as ImageNet and LabelMe, and a novel “cross-domain ” label propagation approach. Our proposed method consists of two major novel components, Exemplar SVM based label propagation, which effectively addresses the cross-domain issue, and a graphical model based contextual refinement incorporating 3D constraints. Most importantly, the entire process does not require any training data from the target scenes, also with good scalability towards large scale applications. We evaluate our approach on the well-known Cornell Point Cloud Dataset, achieving much greater efficiency and comparable accuracy even without any 3D training data. Our approach shows further major gains in accuracy when the training data from the target scenes is used, outperforming state-ofthe-art approaches with far better efficiency.

5 0.10115518 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 0.096213862 178 cvpr-2013-From Local Similarity to Global Coding: An Application to Image Classification

7 0.094792932 236 cvpr-2013-K-Means Hashing: An Affinity-Preserving Quantization Method for Learning Binary Compact Codes

8 0.092659213 268 cvpr-2013-Leveraging Structure from Motion to Learn Discriminative Codebooks for Scalable Landmark Classification

9 0.089357436 166 cvpr-2013-Fast Image Super-Resolution Based on In-Place Example Regression

10 0.084543504 69 cvpr-2013-Boosting Binary Keypoint Descriptors

11 0.077473387 393 cvpr-2013-Separating Signal from Noise Using Patch Recurrence across Scales

12 0.076087616 260 cvpr-2013-Learning and Calibrating Per-Location Classifiers for Visual Place Recognition

13 0.073788732 67 cvpr-2013-Blocks That Shout: Distinctive Parts for Scene Classification

14 0.068656899 59 cvpr-2013-Better Exploiting Motion for Better Action Recognition

15 0.068417102 234 cvpr-2013-Joint Spectral Correspondence for Disparate Image Matching

16 0.066608913 343 cvpr-2013-Query Adaptive Similarity for Large Scale Object Retrieval

17 0.065129273 246 cvpr-2013-Learning Binary Codes for High-Dimensional Data Using Bilinear Projections

18 0.064729668 82 cvpr-2013-Class Generative Models Based on Feature Regression for Pose Estimation of Object Categories

19 0.061251648 304 cvpr-2013-Multipath Sparse Coding Using Hierarchical Matching Pursuit

20 0.060908198 248 cvpr-2013-Learning Collections of Part Models for Object Recognition

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.133), (1, -0.033), (2, -0.019), (3, 0.031), (4, 0.023), (5, 0.01), (6, -0.035), (7, -0.05), (8, -0.091), (9, -0.056), (10, -0.053), (11, -0.003), (12, 0.057), (13, 0.018), (14, 0.059), (15, -0.041), (16, 0.005), (17, -0.014), (18, 0.077), (19, -0.015), (20, 0.063), (21, 0.03), (22, 0.048), (23, -0.077), (24, -0.016), (25, 0.116), (26, -0.045), (27, -0.036), (28, -0.021), (29, -0.042), (30, 0.017), (31, 0.018), (32, 0.045), (33, 0.032), (34, -0.046), (35, -0.026), (36, 0.003), (37, 0.001), (38, 0.063), (39, -0.006), (40, 0.067), (41, 0.039), (42, 0.08), (43, 0.008), (44, 0.041), (45, 0.013), (46, 0.018), (47, -0.046), (48, 0.02), (49, -0.041)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.95913208 404 cvpr-2013-Sparse Quantization for Patch Description

Author: Xavier Boix, Michael Gygli, Gemma Roig, Luc Van_Gool

2 0.74366862 69 cvpr-2013-Boosting Binary Keypoint Descriptors

Author: Tomasz Trzcinski, Mario Christoudias, Pascal Fua, Vincent Lepetit

Abstract: Binary keypoint descriptors provide an efficient alternative to their floating-point competitors as they enable faster processing while requiring less memory. In this paper, we propose a novel framework to learn an extremely compact binary descriptor we call BinBoost that is very robust to illumination and viewpoint changes. Each bit of our descriptor is computed with a boosted binary hash function, and we show how to efficiently optimize the different hash functions so that they complement each other, which is key to compactness and robustness. The hash functions rely on weak learners that are applied directly to the imagepatches, whichfrees usfrom any intermediate representation and lets us automatically learn the image gradient pooling configuration of the final descriptor. Our resulting descriptor significantly outperforms the state-of-the-art binary descriptors and performs similarly to the best floating-point descriptors at a fraction of the matching time and memory footprint.

3 0.71505588 246 cvpr-2013-Learning Binary Codes for High-Dimensional Data Using Bilinear Projections

Author: Yunchao Gong, Sanjiv Kumar, Henry A. Rowley, Svetlana Lazebnik

Abstract: Recent advances in visual recognition indicate that to achieve good retrieval and classification accuracy on largescale datasets like ImageNet, extremely high-dimensional visual descriptors, e.g., Fisher Vectors, are needed. We present a novel method for converting such descriptors to compact similarity-preserving binary codes that exploits their natural matrix structure to reduce their dimensionality using compact bilinear projections instead of a single large projection matrix. This method achieves comparable retrieval and classification accuracy to the original descriptors and to the state-of-the-art Product Quantization approach while having orders ofmagnitudefaster code generation time and smaller memory footprint.

4 0.68403995 79 cvpr-2013-Cartesian K-Means

Author: Mohammad Norouzi, David J. Fleet

5 0.68114674 38 cvpr-2013-All About VLAD

Author: unkown-author

Abstract: The objective of this paper is large scale object instance retrieval, given a query image. A starting point of such systems is feature detection and description, for example using SIFT. The focus of this paper, however, is towards very large scale retrieval where, due to storage requirements, very compact image descriptors are required and no information about the original SIFT descriptors can be accessed directly at run time. We start from VLAD, the state-of-the art compact descriptor introduced by J´ egou et al. [8] for this purpose, and make three novel contributions: first, we show that a simple change to the normalization method significantly improves retrieval performance; second, we show that vocabulary adaptation can substantially alleviate problems caused when images are added to the dataset after initial vocabulary learning. These two methods set a new stateof-the-art over all benchmarks investigated here for both mid-dimensional (20k-D to 30k-D) and small (128-D) descriptors. Our third contribution is a multiple spatial VLAD representation, MultiVLAD, that allows the retrieval and local- ization of objects that only extend over a small part of an image (again without requiring use of the original image SIFT descriptors).

6 0.67952609 421 cvpr-2013-Supervised Kernel Descriptors for Visual Recognition

7 0.67470276 304 cvpr-2013-Multipath Sparse Coding Using Hierarchical Matching Pursuit

8 0.66718084 268 cvpr-2013-Leveraging Structure from Motion to Learn Discriminative Codebooks for Scalable Landmark Classification

9 0.6428889 53 cvpr-2013-BFO Meets HOG: Feature Extraction Based on Histograms of Oriented p.d.f. Gradients for Image Classification

10 0.6099363 319 cvpr-2013-Optimized Product Quantization for Approximate Nearest Neighbor Search

11 0.56806219 178 cvpr-2013-From Local Similarity to Global Coding: An Application to Image Classification

12 0.55447406 166 cvpr-2013-Fast Image Super-Resolution Based on In-Place Example Regression

13 0.54252076 369 cvpr-2013-Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination

14 0.53260911 275 cvpr-2013-Lp-Norm IDF for Large Scale Image Search

15 0.53086388 464 cvpr-2013-What Makes a Patch Distinct?

16 0.52199268 200 cvpr-2013-Harvesting Mid-level Visual Concepts from Large-Scale Internet Images

17 0.51908076 169 cvpr-2013-Fast Patch-Based Denoising Using Approximated Patch Geodesic Paths

18 0.51467568 393 cvpr-2013-Separating Signal from Noise Using Patch Recurrence across Scales

19 0.51196116 266 cvpr-2013-Learning without Human Scores for Blind Image Quality Assessment

20 0.49788651 371 cvpr-2013-SCaLE: Supervised and Cascaded Laplacian Eigenmaps for Visual Object Recognition Based on Nearest Neighbors

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(10, 0.121), (16, 0.019), (26, 0.018), (28, 0.014), (33, 0.271), (35, 0.307), (67, 0.079), (69, 0.035), (87, 0.053)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.81714296 404 cvpr-2013-Sparse Quantization for Patch Description

Author: Xavier Boix, Michael Gygli, Gemma Roig, Luc Van_Gool

2 0.76749194 366 cvpr-2013-Robust Region Grouping via Internal Patch Statistics

Author: Xiaobai Liu, Liang Lin, Alan L. Yuille

Abstract: In this work, we present an efficient multi-scale low-rank representation for image segmentation. Our method begins with partitioning the input images into a set of superpixels, followed by seeking the optimal superpixel-pair affinity matrix, both of which are performed at multiple scales of the input images. Since low-level superpixel features are usually corrupted by image noises, we propose to infer the low-rank refined affinity matrix. The inference is guided by two observations on natural images. First, looking into a single image, local small-size image patterns tend to recur frequently within the same semantic region, but may not appear in semantically different regions. We call this internal image statistics as replication prior, and quantitatively justify it on real image databases. Second, the affinity matrices at different scales should be consistently solved, which leads to the cross-scale consistency constraint. We formulate these two purposes with one unified formulation and develop an efficient optimization procedure. Our experiments demonstrate the presented method can substantially improve segmentation accuracy.

3 0.7405166 53 cvpr-2013-BFO Meets HOG: Feature Extraction Based on Histograms of Oriented p.d.f. Gradients for Image Classification

Author: Takumi Kobayashi

Abstract: Image classification methods have been significantly developed in the last decade. Most methods stem from bagof-features (BoF) approach and it is recently extended to a vector aggregation model, such as using Fisher kernels. In this paper, we propose a novel feature extraction method for image classification. Following the BoF approach, a plenty of local descriptors are first extracted in an image and the proposed method is built upon the probability density function (p.d.f) formed by those descriptors. Since the p.d.f essentially represents the image, we extract the features from the p.d.f by means of the gradients on the p.d.f. The gradients, especially their orientations, effectively characterize the shape of the p.d.f from the geometrical viewpoint. We construct the features by the histogram of the oriented p.d.f gradients via orientation coding followed by aggregation of the orientation codes. The proposed image features, imposing no specific assumption on the targets, are so general as to be applicable to any kinds of tasks regarding image classifications. In the experiments on object recog- nition and scene classification using various datasets, the proposed method exhibits superior performances compared to the other existing methods.

4 0.71252877 408 cvpr-2013-Spatiotemporal Deformable Part Models for Action Detection

Author: Yicong Tian, Rahul Sukthankar, Mubarak Shah

Abstract: Deformable part models have achieved impressive performance for object detection, even on difficult image datasets. This paper explores the generalization of deformable part models from 2D images to 3D spatiotemporal volumes to better study their effectiveness for action detection in video. Actions are treated as spatiotemporal patterns and a deformable part model is generated for each action from a collection of examples. For each action model, the most discriminative 3D subvolumes are automatically selected as parts and the spatiotemporal relations between their locations are learned. By focusing on the most distinctive parts of each action, our models adapt to intra-class variation and show robustness to clutter. Extensive experiments on several video datasets demonstrate the strength of spatiotemporal DPMs for classifying and localizing actions.

5 0.7116999 225 cvpr-2013-Integrating Grammar and Segmentation for Human Pose Estimation

Author: Brandon Rothrock, Seyoung Park, Song-Chun Zhu

Abstract: In this paper we present a compositional and-or graph grammar model for human pose estimation. Our model has three distinguishing features: (i) large appearance differences between people are handled compositionally by allowingparts or collections ofparts to be substituted with alternative variants, (ii) each variant is a sub-model that can define its own articulated geometry and context-sensitive compatibility with neighboring part variants, and (iii) background region segmentation is incorporated into the part appearance models to better estimate the contrast of a part region from its surroundings, and improve resilience to background clutter. The resulting integrated framework is trained discriminatively in a max-margin framework using an efficient and exact inference algorithm. We present experimental evaluation of our model on two popular datasets, and show performance improvements over the state-of-art on both benchmarks.

6 0.71165657 60 cvpr-2013-Beyond Physical Connections: Tree Models in Human Pose Estimation

7 0.71130013 248 cvpr-2013-Learning Collections of Part Models for Object Recognition

8 0.71086562 387 cvpr-2013-Semi-supervised Domain Adaptation with Instance Constraints

9 0.71044588 14 cvpr-2013-A Joint Model for 2D and 3D Pose Estimation from a Single Image

10 0.71020609 414 cvpr-2013-Structure Preserving Object Tracking

11 0.71016628 122 cvpr-2013-Detection Evolution with Multi-order Contextual Co-occurrence

12 0.71010005 325 cvpr-2013-Part Discovery from Partial Correspondence

13 0.70992041 206 cvpr-2013-Human Pose Estimation Using Body Parts Dependent Joint Regressors

14 0.70970345 446 cvpr-2013-Understanding Indoor Scenes Using 3D Geometric Phrases

15 0.70919895 419 cvpr-2013-Subspace Interpolation via Dictionary Learning for Unsupervised Domain Adaptation

16 0.70896924 256 cvpr-2013-Learning Structured Hough Voting for Joint Object Detection and Occlusion Reasoning

17 0.7085138 249 cvpr-2013-Learning Compact Binary Codes for Visual Tracking

18 0.70843136 221 cvpr-2013-Incorporating Structural Alternatives and Sharing into Hierarchy for Multiclass Object Recognition and Detection

19 0.70830297 314 cvpr-2013-Online Object Tracking: A Benchmark

20 0.70824814 339 cvpr-2013-Probabilistic Graphlet Cut: Exploiting Spatial Structure Cue for Weakly Supervised Image Segmentation