Multi-scale and Discriminative Part Detectors Based Features for Multi-label Image Classification

Size: px
Start display at page:

Download "Multi-scale and Discriminative Part Detectors Based Features for Multi-label Image Classification"

Transcription

1 Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) Mult-sale and Dsrmnatve Part Detetors Based Features for Mult-lael Image Classfaton Gong Cheng, Deheng Gao, Yang Lu, Junwe Han * Shool of Automaton, orthwestern Polytehnal Unversty, X'an, Chna State ey Laoratory of Integrated Serves etworks, Xdan Unversty, X'an, Chna {henggong9, xdanluyang, junwehan00}@gmal.om, deheng@mal.nwpu.edu.n Astrat Convolutonal neural networks (Cs) have shown ther promse for mage lassfaton task. However, gloal C features stll lak geometr nvarane for addressng the prolem of ntra-lass varatons and so are not optmal for mult-lael mage lassfaton. hs paper proposes a new and effetve framework ult upon Cs to learn Mult-sale and Dsrmnatve Part Detetors (MsDPD)-ased feature representatons for mult-lael mage lassfaton. Spefally, at eah sale level, we () frst present an entropy-rank ased sheme to generate and selet a set of dsrmnatve part detetors (DPD), and then () otan a numer of DPD-ased onvolutonal feature maps wth eah feature map representng the ourrene proalty of a partular part detetor and learn DPD-ased features y usng a task-drven poolng sheme. he two steps are formulated nto a unfed framework y developng a new ojetve funton, whh jontly trans part detetors nrementally and ntegrates the learnng of feature representatons nto the lassfaton task. Fnally, the mult-sale features are fused to produe the predtons. Expermental results on PASCAL VOC 007 and VOC 0 datasets demonstrate that the proposed method aheves etter auray when ompared wth the exstng state-of-the-art mult-lael lassfaton methods. Introduton Mult-lael mage lassfaton has attrated partular attenton reently drven y ts road applatons [Geng and Luo, 04; George and Floerkemeer, 04; Gong et al., 03; Jng et al., 05; L et al., 06a; L et al., 06; L et al., 07; Murthy et al., 06; an et al., 05; Wang et al., 06; We et al., 04; We et al., 06; Xe et al., 07; Yeh et al., 07; Zhu et al., 07]. he task of mult-lael mage lassfaton s to predt the presene or asene of multple spef ojet ategores n an mage. Compared wth sngle-lael mage lassfaton whh has een atvely stud- * Correspondng author. Fgure : Mult-lael mages from the PASCAL VOC 007 dataset. he ntra-lass varatons and the omposton and nteraton etween dfferent ojet ategores make the task of mult-lael mage lassfaton more hallengng. ed n reent years [Herranz et al., 06; rzhevsky et al., 0; Smon et al., 04; Smonyan and Zsserman, 05; Szegedy et al., 05], mult-lael mage lassfaton s a more pratal prolem eause most of the real-world mages usually ontan multple ojets from dfferent ategores. Besdes, as shown n Fgure, eah ojet lass n real-world mult-lael mages often has large ntra-lass varatons aused y oluson, sale, vewpont, llumnaton, et., and the omposton and nteraton etween ojet ategores also nrease the omplexty of the prolem, whh make the task of mult-lael mage lassfaton more hallengng. Durng the past few years, varous deep learnng methods espeally onvolutonal neural networks (Cs) have shown ther promse as a unversal representaton and have domnated most of the reent works on mage lassfaton task. However, most researh efforts made on mage lassfaton manly fous on addressng the task of sngle-lael mage lassfaton. Although several reent works [Oqua et al., 04; Sharf Razavan et al., 04; Smonyan and Zsserman, 05] have demonstrated that a pre-traned C model an also e straghtforwardly transferred to mult-lael mage lassfaton, they do not perform well for reognzng omplex ojet layouts and senes n mult-lael mages. hs manly eause gloal C features stll lak geometr nvarane for addressng the prolem of ntra-lass varatons and so are not optmal for mult-lael mage lassfaton. 649

2 Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) Fgure : he arhteture of the proposed MsDPD-ased mult-lael mage lassfaton framework. At eah sale level (denoted y gray loks), the DPD-ased feature representatons are learned from the C onvolutonal features y usng our proposed optmzaton method, as shown n Fgure 3. he ultmate mult-lael predtons are otaned y aggregatng the features from dfferent sale levels. Fgure 3: he proposed unfed optmzaton framework for the jont tranng of dsrmnatve part detetors (DPD) and DPD-ased feature representatons. Spefally, for a onvolutonal layer of sze m m wth n hannels, we onvolve t wth n part detetors to produe a numer of part detetors-ased feature maps of sze m m wth hannels, followed y a task-drven poolng step to produe the fnal -dmensonal DPD-ased feature representaton. In ths paper, we propose a novel and effetve framework ult upon Cs to learn mult-sale and dsrmnatve part detetors (MsDPD)-ased feature representatons for the task of mult-lael mage lassfaton, as shown n Fgure. Spefally, at eah sale, we frst present an ojet-proposal-free and entropy-rank ased sheme to generate and selet a numer of dsrmnatve part detetors (DPD). hen, we otan a set of DPD-ased feature maps wth eah feature map representng the ourrene proalty of a partular part detetor, and learn the pooled DPD-ased features y usng a task-drven poolng sheme. We formulate the two steps nto a unfed optmzaton framework, whh trans part detetors nrementally and ntegrates the learnng of feature representatons nto the lassfaton task, as shown n Fgure 3. Fnally, the features from dfferent sale levels are aggregated to produe the ultmate mult-lael predtons. In the experments, we evaluate the proposed framework on the PASCAL VOC 007 and VOC 0 datasets [Everngham et al., 05] and aheve state-of-the-art results when ompared wth the exstng mult-lael mage lassfaton methods. o sum up, our man ontrutons are as follows. Frst, we propose a unfed framework y leveragng the hghly expressve Cs to learn a knd of dsrmnatve part detetors-ased feature representaton, termed MsDPD, to address the prolems of ntra-lass varatons faed for mult-lael mage lassfaton. he proposed approah formulates the tranng of part detetors and the learnng of feature representatons nto a unfed optmzaton framework y developng a new ojetve funton. Seond, we present an entropy-rank ased sheme to evaluate the dstntveness of part detetors and then tran part detetors nrementally y mnng relale nstanes teratvely. hrd, we propose a task-drven poolng tehnque to ntegrate the learnng of feature representaton nto lassfaton task to mprove ts generalty. 650

3 Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) Fourth, dfferent from prevous regon proposal-ased mage lassfaton methods [We et al., 06; Wu et al., 05; Yang et al., 06], our method does not need ground-truth oundng oxes or ojet proposals, makng the proposed method more effent and pratal. We have onfrmed through experments that the feature representaton otaned y usng the proposed method s apale of delverng state-of-the-art results on two popular mult-lael lassfaton enhmarks nludng PASCAL VOC 007 and VOC 0 datasets. Methodology Whle many C-ased methods have aheved suessful results on mage lassfaton, most of them are developed for sngle-lael mage lassfaton y extratng gloal C features. Inspred y the fat that eah ojet lass n mult-lael mages generally exhts dramatally dfferent appearanes, shapes, olusons and nteratons, we propose to extrat dsrmnatve part detetors-ased features, a knd of loal C-ased features, to handle the prolem of ntra-lass varatons. Here, part detetors are used to apture generalzed ojets and ther parts that are dsrmnatve (eng dfferent enough from eah other) and representatve (ourng frequently enough). As shown n Fgure, the ore task of ths proposed method s to learn Mult-sale and Dsrmnatve Part Detetors (MsDPD)-ased features for mult-lael mage lassfaton. Fgure 3 llustrates how to learn DPD-ased features from eah sale of onvolutonal features denoted y gray loks. Spefally, we frst present an entropy-rank ased sheme to generate a numer of dsrmnatve part detetors. hen, we otan part detetors-ased onvolutonal feature maps and generate the pooled feature representatons y usng a task-drven feature poolng sheme. For ease of optmzaton we ntegrate the two steps nto a unfed framework y developng a new ojetve funton to jontly tran part detetors and learn the feature representatons. he fnal mult-lael predton results are otaned y fusng the features from dfferent sale levels.. Model Arhteture Fgure llustrates the overall arhteture of our MsDPD framework. he as onfguraton of our model s smlar to that of Sngle Shot MultBox Detetor (SSD) [Lu et al., 06]. he early layers (Conv to Conv7) are transferred from the pre-traned VGGet-6 [Smonyan and Zsserman, 05], where the onvolutonal layers Conv6 and Conv7 are onverted from the fully-onneted layers FC6 and FC7 y usng a sheme that s smlar to SSD as follows: susample parameters from FC6 and FC7, hange pool5 from - s to s, and use the `a trous algorthm to fll the "holes". he fully-onneted layers are onverted to onvolutonal ones to ope wth the unertanty for the loalzaton of ojet parts. hese layers are followed y some extra onvolutonal layers (Conv8 to Conv) to extrat muh deeper features and even gger ojet parts. he last onvolutonal layer (Conv_) s used to fne-tune the network y usng mult-lael mages. he detaled model parameters an e found n Fgure. In ths work, DPD-ased feature representatons are extrated ased on four sales of onvolutonal layers nludng Conv7, Conv8_, Conv9_, and Conv0_ (denoted y gray loks n Fgure ), whh derease n sze progressvely to allow the detetons of ojet parts at multple sales. Spefally, for a onvolutonal layer of sze m m wth n hannels, we onvolve t wth n part detetors to produe a numer of part detetors-ased onvolutonal feature maps of sze m m wth hannels, where eah feature map represents the ourrene proalty of a partular part detetor, followed y a task-drven poolng step to produe the fnal -dmensonal DPD-ased feature representaton. For ease of referene, we ndex the four DPD-ased feature layers and the last onvolutonal layer y usng sale through sale 5. ext we desre how to tran part detetors and learn part detetors-ased feature representatons.. Intalzng Dsrmnatve Part Detetors (DPD) o ntalze the anddate part detetors that are shared aross all mage ategores, we randomly sample a large numer of (aout one hundred thousands) pxels from eah sale of the onvolutonal feature maps of all tranng mages. Eah pxel from the feature maps an e onsdered as a loal C feature whh has a very large reeptve feld n the orgnal mage, and the length of the pxel equals the hannel numer of the onvolutonal features. hen we perform k-means lusterng over these sampled pxels and only retan suffently large lusters to ensure the representatveness, where eah luster orresponds to a to-e-learned anddate part detetor. We onsder an ojet part s dsrmnatve f t only appears frequently n some spef mage ategores rather than almost all lasses. For example, "wheel" wll our n the ojet lasses of "us" and "ar", so the entropy would e low. In ontrast, a non-dsrmnatve "sky" ould our unformly n almost any of the lasses wth hgher entropy. o selet dsrmnatve part detetors, we ntrodue an entropy-rank ased sheme to measure the dsrmnaton of eah part detetor y omputng ther entropes aross all mage ategores. Spefally, the entropy ED for a part detetor D s omputed y C E D p D p D () log where C s the numer of mage lasses and p D s the fraton of the memers of part detetor D that are from the mages of the -th lass. hen, we take the entropy as a measure of dsrmnaton of a part detetor to selet detetors wth low entropy values..3 ranng DPD and Learnng DPD-ased Feature Let { X,,, } e the set of tranng mages and { Y,,, } e the set of mage laels of, where X { n xj j,,, M} s represented y the entres from ts C onvolutonal layer of sze m m wth n hannels and M m, s the total numer of tranng C mages, Y denotes the ground truth lael vetor of sample X wth at least one element eng, and C s the numer of mage lasses. For a gven tranng mage X, let 65

4 O X { O xj j,,, M} e ts DPD-ased feature maps, X e ts pooled DPD-ased feature, P X e ts mult-lael predton result, W, e the parameters of the to-e-learned DPD, and W, e the n parameters of C mult-lael lassfers, where W,, C C W, and. hus, O x j and P an e omputed y X j S W j O x x () P X W (3) x x x and where S( ) exp( )/ exp( ) ( x) exp( x ) are the softmax and sgmod non-lnear atvaton funtons, whh are used for predtng the ourrene proaltes of DPD and the mult-lael lassfaton results, respetvely. As shown n Fgure 3, to leverage Cs to learn effetve feature representaton, we formulate the tranng of DPD and the learnng of DPD-ased feature representaton nto a unfed framework, whh jontly trans part detetors nrementally and ntegrates DPD-ased feature learnng nto lassfaton. o ths end, we develop a new ojetve funton as follows, whh ontans three terms nludng an mage-level lassfaton loss term, a generalzed max poolng regularzaton term, and an ojet part-level lassfaton loss term: J J mn J3W,,, W,, J W,, where and are two trade-off parameters that ontrol the relatve mportane of these three terms. ) Image-level lassfaton loss term. hs term s defned as sgmod ross-entropy loss funton for mult-lael mage lassfaton. It ams to mnmze the lassfaton error for the gven tranng mages and s omputed y C Y log P X Y log PX J where Y and P X denote the -th entres of Y and P X, respetvely. ) Generalzed max poolng regularzaton term. hs term s used to learn the pooled DPD-ased feature X from the nput OX y enforng the pooled representaton to e lose to eah olumn of the nput OX, whh s omputed y usng the followng formula M J Oxj X X (6) M j Smlar to [Murray and Perronnn, 04; Xe et al., 05], y usng ths poolng regularzaton term, the learned DPD-ased feature ould enfore the dot produt smlarty etween j O x and the pooled feature X to e a onstant one. By ntegratng feature learnng nto lassfaton, we an use O X to get a task-drven feature more nformaton from Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) (4) (5) representaton whh s more sutale for lassfaton than tradtonal poolng strateges suh as max/average-poolng. 3) Ojet part-level lassfaton loss term. hs term s defned as softmax ross-entropy loss funton for ojet part lassfaton. It ams to mnmze the lassfaton error for the mned ojet part nstanes and s omputed y t J3 yl log l k k t O x (7) l k where y l stands for ojet part lael vetor of ojet part nstane x l wth only one element eng, s the numer of DPD seleted n suseton., and t s the numer of hgh-onfdent ojet parts we selet n eah teraton used for updatng eah part detetor. In suh way, we an mne relale nstanes teratvely and tran the DPD nrementally..4 Optmzaton o solve the optmzaton prolem of Eq. (4), we present a smple EM-lke teratve mnmzaton method to update W,, W, and alternatvely va stohast gradent desent method (SGD) [Wllams and Hnton, 986]. ) Intalzaton. Gven seleted part detetors, we ntalze the parameters W, y usng Eq. (7). he DPD-ased feature representaton for all mages are ntalzed y usng (6) and generalzed max poolng method [Murray and Perronnn, 04]. he parameters W, are ntalzed y usng Eq. (5). ) Updatng W, and W, y fxng. he gradents of the ojetve funton J wth respet to the parameters, W, an e omputed y J W W and M t g where M (8) j xz j j xl O xl yl j t l J g M M t j zj O xl yl j t l (9) J XPXY (0) W J PXY g j and z j are defned as follows g X j j () O x () X g z O x O x (3) j j j j wth the operaton denotng element-wse multplaton. hus, the parameters W, and W, an e updated y usng gradent desent method as follows W W J W, J (4) 65

5 Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) W W J W, J where s the learnng rate. 3) Updatng y fxng W, and W, (5). he gradents of the ojetve funton J wth respet to an e omputed y J X W PXY M M O x j j X Oxj X hus, the parameter an e updated as follows X X J X (6) (7).5 Mult-lael Image Classfaton After optmzng Eq. (4), the pooled DPD-ased features for all tranng samples are learned at the same tme. However, the feature representatons of the test mages stll need to e learned. Sne the part detetors make the dstrutons of tranng and test data onsstent, we an otan DPD-ased features of test mages y optmzng Eq. (6). For the ultmate predtons, we onatenate the features from dfferent sale levels to tran a set of sgmod lassfers for predton. 3 Experments In the experments, we evaluate our method on PASCAL VOC 007 and VOC 0 datasets [Everngham et al., 05], whh have een wdely used for mult-lael mage lassfaton y predtng whether the ojet s present/asent n the mage. he performane s measured y usng the average preson (AP) and the mean AP over all ojet lasses. 3. Parameter Settngs We tran the proposed model as shown n Fgure y usng SGD wth ntal learnng rate of 0-4 for the early layers (Conv to Conv7), ntal learnng rate of 0-3 for the latter layers (Conv8 to Conv), momentum of 0.9, weght deay of , and ath sze of 3. he learnng rate deays y 0. after 60k teratons and s fxed for the rest 0k teratons. For the tranng of DPD and DPD-ased features, the parameters n Eq. (4) are set to = and =0.0, and the learnng rate n Eqs. (4), (5), (7) s set to 0.0. For the ntalzaton of DPD, we run k-means lusterng on the onvolutonal feature maps of sales to 4 wth the luster numers eng set to 000, 000, 400, and 00, and then take the entropy-rank ased sheme as a measure to selet 700, 400, 300, and 00 detetors, respetvely. 3. Expermental Results Comparson of features from dfferent sales. We frst gve the results otaned y usng dfferent sales on PAS- CAL VOC 007 dataset. ale reports the detaled results. As shown n ale, some ojet lasses, suh as "rd" and "ottle", fre on small sales and some ojet ategores, suh as "person" and "tran" fre on g sales. hs s eause that our MsDPD feature layers are dereased n sze progressvely to allow the predtons of ojets and ther parts at multple sales, therey for etter apture of ojet varatons aused y vewpont, sale, oluson, et. he est results are otaned y fusng the features of dfferent sales. State-of-the-art C-ased methods. he followng C-ased methods are used for omparson: VGG-6-SVM and VGG-9-SVM [Smonyan and Zsserman, 05], Res- et-0-sgmod [He et al., 06], SDE [Xe et al., 07a], HCP [We et al., 06], C-R [Wang et al., 06], and FeV+LV-0-VD [Yang et al., 06]. he work of [Smonyan and Zsserman, 05] densely extrats 4096-D C features aross fve mage sales {56,384,5,640,748} of the gven mage wth VGG-6 and VGG-9, performs gloal average poolng on the resultng C features, and fnally lassfes the mage wth lnear SVM lassfers. Reset-0-Sgmod trans a mult-lael lassfaton system usng a pre-traned Reset-0 model [He et al., 06] wth a sgmod ross entropy loss funton, densely omputes sgmod outputs aross fve mage sales {56,384,5,640,748} of the gven mage, and fnally performs lassfaton y max-poolng the resultng sgmod outputs as HCP [We et al., 06]. SDE [Xe et al., 07a] presented a feature learnng framework y optmzng the features wth the am of learnng seletve, dsrmnatve and equalzng representatons. HCP [We et al., 06] proposed to address the mult-lael lassfaton y extratng ojet proposals from the gven mages and the fnal mage-level sores are otaned y max-poolng the sores of the proposals. C-R [Wang et al., 06] omned Rs wth Cs n a unfed framework to learn a jont mage-lael emeddng. FeV+LV-0-VD [Yang et al., 06] proposed a mult-vew mult-nstane framework to utlze oth weak and strong laels (oundng ox). Comparson wth state-of-the-art methods on PASCAL VOC 007 dataset. ale summarzes the results of our MsDPD method and the aforementoned seven state-of-the-art methods on PASCAL VOC 007 dataset. As shown n ale : () Compared wth gloal C-ased approahes suh as VGG-6-SVM and VGG-9-SVM [Smonyan and Zsserman, 05], Reset-0-Sgmod [He et al., 06], and C-R [Wang et al., 06], our proposed method otans sgnfant performane gans of 4.%, 3.3% and 9.5% n terms of map. hs shows the superorty of our loal C-ased method. () Compared wth other methods, suh as HCP [We et al., 06], SDE [Xe et al., 07a], and FeV+LV-0-VD [Yang et al., 06], whh an e regarded as a knd of loal feature ased methods, our method stll outperforms them wth a g margn measured n terms of map (at least.3%). Comparson wth state-of-the-art methods on PASCAL VOC 0 dataset. We report our expermental results n ale 3 and ompare t wth sx state-of-the-art C-ased methods on VOC 0 dataset. he results are onsstent wth those on the VOC 007 dataset. o e spef, we aheve state-of-the-art results for 6 out of 0 ojet ategores. Espeally for the dffult ategores suh as "har", "ow", 653

6 Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) Method aero ke rd oat ottle us ar at har ow tale dog horse mke person plant sheep sofa tran tv map Sale Sale Sale Sale Sale MsDPD ale : Classfaton results (%) on the PASCAL VOC 007 test set otaned y usng dfferent sales of MsDPD and ther fuson (sales to 5). he entres wth the est APs for eah ojet ategory are old-faed. Method aero ke rd oat ottle us ar at har ow tale dog horse mke person plant sheep sofa tran tv map VGG-6-SVM MS VGG-9-SVM MS Reset-0-Sgmod MS HCP C-R FeV+LV-0-VD SDE Our MsDPD method ale : Classfaton results (%) on the PASCAL VOC 007 test set otaned y usng state-of-the-art C-ased methods and our proposed MsDPD method. : MS denotes the results otaned y usng a mult-sale sheme wth fve mage sales {56,384,5,640,748}. Method aero ke rd oat ottle us ar at har ow tale dog horse mke person plant sheep sofa tran tv map VGG-6-SVM MS VGG-9-SVM MS Reset-0-Sgmod MS HCP FeV+LV-0-VD SDE Our MsDPD method ale 3: Classfaton results (%) on the PASCAL VOC 0 test set otaned y usng state-of-the-art C-ased methods and our proposed MsDPD method. : MS denotes the results otaned y usng a mult-sale sheme wth fve mage sales {56,384,5,640,748}. "tale", "plant", and "sofa", our method shows good performane. hs sgnfant performane gan shows the effetveness of our DPD-ased feature representaton. Alaton experments. o analyze the mportane of eah omponent of our method (part detetors and task-drven poolng), we onduted alaton experments on the PAS- CAL VOC 007 dataset. ale 4 shows the results otaned wth part detetors and wthout part detetors (pool features from Conv7/Conv8_/Conv9_/Conv0_ layers) y usng dfferent poolng strateges measured n terms of map. As shown n ale 4, task-drven poolng otans the hghest map than max-poolng and average-poolng. More mportantly, y usng our proposed part detetors ould otan g auray gans ompared wth that otaned y dretly poolng features from the orgnal onvolutonal layers. 4 Conluson In ths paper, we proposed to uld upon Cs to learn part detetors-ased features for mult-lael mage lassfaton. o ths end, we frst present an entropy-rank ased sheme to Max-poolng 89. Wthout part detetors Average-poolng 88. ask-drven poolng 90.7 Max-poolng 93. Wth part detetors Average-poolng 87.9 ask-drven poolng 93.5 ale 4: Alaton expermental results (map, %) on the PASCAL VOC 007 test set otaned wth part detetors and wthout part detetors y usng dfferent poolng strateges. otan a set of dsrmnatve part detetors. hen, we generate part detetors-ased onvolutonal feature maps and learn part detetors-ased features wth a task-drven poolng sheme. For optmzaton, the aforementoned two steps are formulated nto a unfed framework y developng a new ojetve funton, whh nrementally trans part detetors and ntegrates the learnng of feature representatons nto the lassfaton task. However, y usng the proposed ojetve funton t s dffult to tran the whole network end-to-end. Our future work wll address ths ssue. 654

7 Proeedngs of the wenty-seventh Internatonal Jont Conferene on Artfal Intellgene (IJCAI-8) Aknowledgments hs work was supported n part y the SFC under Grants and 64733, n part y the atural Sene Bas Researh Plan n Shaanx Provne of Chna under Grant 07JM6044, and n part y the Fundamental Researh Funds for the Central Unverstes under Grant 3008zy03. Referenes [Everngham et al., 05] M. Everngham, S. A. Eslam, L. Van Gool, C.. Wllams, J. Wnn, and A. Zsserman. he pasal vsual ojet lasses hallenge: A retrospetve. IJCV, (): 98-36, 05. [Geng and Luo, 04] X. Geng and L. Luo. Multlael rankng wth nonsstent rankers. In CVPR, 04. [George and Floerkemeer, 04] M. George and C. Floerkemeer. Reognzng produts: A per-exemplar mult-lael mage lassfaton approah. In ECCV, 04. [Gong et al., 03] Y. Gong, Y. Ja,. Leung, A. oshev, and S. Ioffe. Deep onvolutonal rankng for multlael mage annotaton. arxv preprnt arxv:3.4894, 03. [He et al., 06]. He, X. Zhang, S. Ren, and J. Sun. Deep resdual learnng for mage reognton. In CVPR, 06. [Herranz et al., 06] L. Herranz, S. Jang, and X. L. Sene reognton wth Cs: ojets, sales and dataset as. In CVPR, 06. [Jng et al., 05] L. Jng, L. Yang, J. Yu, and M.. g. Sem-supervsed low-rank mappng learnng for mult-lael lassfaton. In CVPR, 05. [rzhevsky et al., 0] A. rzhevsky, I. Sutskever, and G. E. Hnton. Imagenet lassfaton wth deep onvolutonal neural networks. In IPS, 0. [L et al., 06a] C. L, B. Wang, V. Pavlu, and J. Aslam. Condtonal ernoull mxtures for mult-lael lassfaton. In ICML, 06a. [L et al., 06] Q. L, M. Qao, W. Ban, and D. ao. Condtonal graphal lasso for mult-lael mage lassfaton. In CVPR, 06. [L et al., 07] Y. L, Y. Song, and J. Luo. Improvng Parwse Rankng for Mult-lael Image Classfaton. In CVPR, 07. [Lu et al., 06] W. Lu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg. Ssd: Sngle shot multox detetor. In ECCV, 06. [Murray and Perronnn, 04]. Murray and F. Perronnn. Generalzed max poolng. In CVPR, 04. [Murthy et al., 06] V.. Murthy, V. Sngh,. Chen, R. Manmatha, and D. Comanu. Deep deson network for mult-lass mage lassfaton. In CVPR, 06. [Oqua et al., 04] M. Oqua, L. Bottou, I. Laptev, and J. Sv. Learnng and transferrng md-level mage representatons usng onvolutonal neural networks. In CVPR, 04. [Sharf Razavan et al., 04] A. Sharf Razavan, H. Azzpour, J. Sullvan, and S. Carlsson. C features off-the-shelf: an astoundng aselne for reognton. In CVPRW, 04. [Smon et al., 04] M. Smon, E. Rodner, and J. Denzler. Part detetor dsovery n deep onvolutonal neural networks. In ACCV, 04. [Smonyan and Zsserman, 05]. Smonyan and A. Zsserman. Very deep onvolutonal networks for large-sale mage reognton. In ICLR, 05. [Szegedy et al., 05] C. Szegedy, W. Lu, Y. Ja, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhouke, and A. Ranovh. Gong deeper wth onvolutons. In CVPR, 05. [an et al., 05] M. an, Q. Sh, A. van den Hengel, C. Shen, J. Gao, F. Hu, and Z. Zhang. Learnng graph struture for mult-lael mage lassfaton va lque generaton. In CVPR, 05. [Wang et al., 06] J. Wang, Y. Yang, J. Mao, Z. Huang, C. Huang, and W. Xu. C-R: A unfed framework for mult-lael mage lassfaton. In CVPR, 06. [We et al., 04] Y. We, W. Xa, J. Huang, B., J. Dong, Y. Zhao, and S. Yan. C: Sngle-lael to mult-lael. arxv preprnt arxv: , 04. [We et al., 06] Y. We, W. Xa, M. Ln, J. Huang, B., J. Dong, Y. Zhao, and S. Yan. HCP: A flexle C framework for mult-lael mage lassfaton. IEEE PAMI, 38(9): , 06. [Wllams and Hnton, 986] D. Wllams and G. Hnton. Learnng representatons y ak-propagatng errors. ature, 33(6088): , 986. [Wu et al., 05] R. Wu, B. Wang, W. Wang, and Y. Yu. Harvestng dsrmnatve meta ojets wth deep C features for sene lassfaton. In ICCV, 05. [Xe et al., 05] G.-S. Xe, X.-Y. Zhang, X. Shu, S. Yan, and C.-L. Lu. ask-drven feature poolng for mage lassfaton. In ICCV, 05. [Xe et al., 07a] G.-S. Xe, X.-Y. Zhang, S. Yan, and C.-L. Lu. SDE: A ovel Seletve, Dsrmnatve and Equalzng Feature Representaton for Vsual Reognton. IJCV, -4, 07a. [Xe et al., 07] P. Xe, R. Salakhutdnov, L. Mou, and E. P. Xng. Deep Determnantal Pont Proess for Large-Sale Mult-Lael Classfaton. In ICCV, 07. [Yang et al., 06] H. Yang, J. any Zhou, Y. Zhang, B.-B. Gao, J. Wu, and J. Ca. Explot oundng ox annotatons for mult-lael ojet reognton. In CVPR, 06. [Yeh et al., 07] C.-. Yeh, W.-C. Wu, W.-J. o, and Y.-C. F. Wang. Learnng Deep Latent Spae for Mult-Lael Classfaton. In AAAI, 07. [Zhu et al., 07] F. Zhu, H. L, W. Ouyang,. Yu, and X. Wang. Learnng Spatal Regularzaton wth Image-level Supervsons for Mult-lael Image Classfaton. In CVPR,

Matrix-Matrix Multiplication Using Systolic Array Architecture in Bluespec

Matrix-Matrix Multiplication Using Systolic Array Architecture in Bluespec Matrx-Matrx Multplaton Usng Systol Array Arhteture n Bluespe Team SegFault Chatanya Peddawad (EEB096), Aman Goel (EEB087), heera B (EEB090) Ot. 25, 205 Theoretal Bakground. Matrx-Matrx Multplaton on Hardware

More information

Research on Neural Network Model Based on Subtraction Clustering and Its Applications

Research on Neural Network Model Based on Subtraction Clustering and Its Applications Avalable onlne at www.senedret.om Physs Proeda 5 (01 ) 164 1647 01 Internatonal Conferene on Sold State Deves and Materals Sene Researh on Neural Networ Model Based on Subtraton Clusterng and Its Applatons

More information

Multilabel Classification with Meta-level Features

Multilabel Classification with Meta-level Features Multlabel Classfaton wth Meta-level Features Sddharth Gopal Carnege Mellon Unversty Pttsburgh PA 523 sgopal@andrew.mu.edu Ymng Yang Carnege Mellon Unversty Pttsburgh PA 523 ymng@s.mu.edu ABSTRACT Effetve

More information

Performance Evaluation of TreeQ and LVQ Classifiers for Music Information Retrieval

Performance Evaluation of TreeQ and LVQ Classifiers for Music Information Retrieval Performane Evaluaton of TreeQ and LVQ Classfers for Mus Informaton Retreval Matna Charam, Ram Halloush, Sofa Tsekerdou Athens Informaton Tehnology (AIT) 0.8 km Markopoulo Ave. GR - 19002 Peana, Athens,

More information

LOCAL BINARY PATTERNS AND ITS VARIANTS FOR FACE RECOGNITION

LOCAL BINARY PATTERNS AND ITS VARIANTS FOR FACE RECOGNITION IEEE-Internatonal Conferene on Reent Trends n Informaton Tehnology, ICRTIT 211 MIT, Anna Unversty, Chenna. June 3-5, 211 LOCAL BINARY PATTERNS AND ITS VARIANTS FOR FACE RECOGNITION K.Meena #1, Dr.A.Suruland

More information

arxiv: v3 [cs.cv] 31 Oct 2016

arxiv: v3 [cs.cv] 31 Oct 2016 Unversal Correspondene Network Chrstopher B. Choy Stanford Unversty hrshoy@a.stanford.edu JunYoung Gwak Stanford Unversty jgwak@a.stanford.edu Slvo Savarese Stanford Unversty sslvo@stanford.edu arxv:1606.03558v3

More information

Adaptive Class Preserving Representation for Image Classification

Adaptive Class Preserving Representation for Image Classification Adaptve Class Preservng Representaton for Image Classfaton Jan-Xun M,, Qankun Fu,, Wesheng L, Chongqng Key Laboratory of Computatonal Intellgene, Chongqng Unversty of Posts and eleommunatons, Chongqng,

More information

Boosting Weighted Linear Discriminant Analysis

Boosting Weighted Linear Discriminant Analysis . Okada et al. / Internatonal Journal of Advaned Statsts and I&C for Eonoms and Lfe Senes Boostng Weghted Lnear Dsrmnant Analyss azunor Okada, Arturo Flores 2, Marus George Lnguraru 3 Computer Sene Department,

More information

Color Texture Classification using Modified Local Binary Patterns based on Intensity and Color Information

Color Texture Classification using Modified Local Binary Patterns based on Intensity and Color Information Color Texture Classfaton usng Modfed Loal Bnary Patterns based on Intensty and Color Informaton Shvashankar S. Department of Computer Sene Karnatak Unversty, Dharwad-580003 Karnataka,Inda shvashankars@kud.a.n

More information

Computing Cloud Cover Fraction in Satellite Images using Deep Extreme Learning Machine

Computing Cloud Cover Fraction in Satellite Images using Deep Extreme Learning Machine Computng Cloud Cover Fraton n Satellte Images usng Deep Extreme Learnng Mahne L-guo WENG, We-bn KONG, Mn XIA College of Informaton and Control, Nanjng Unversty of Informaton Sene & Tehnology, Nanjng Jangsu

More information

Link Graph Analysis for Adult Images Classification

Link Graph Analysis for Adult Images Classification Lnk Graph Analyss for Adult Images Classfaton Evgeny Khartonov Insttute of Physs and Tehnology, Yandex LLC 90, 6 Lev Tolstoy st., khartonov@yandex-team.ru Anton Slesarev Insttute of Physs and Tehnology,

More information

A MPAA-Based Iterative Clustering Algorithm Augmented by Nearest Neighbors Search for Time-Series Data Streams

A MPAA-Based Iterative Clustering Algorithm Augmented by Nearest Neighbors Search for Time-Series Data Streams A MPAA-Based Iteratve Clusterng Algorthm Augmented by Nearest Neghbors Searh for Tme-Seres Data Streams Jessa Ln 1, Mha Vlahos 1, Eamonn Keogh 1, Dmtros Gunopulos 1, Janwe Lu 2, Shouan Yu 2, and Jan Le

More information

Gabor-Filtering-Based Completed Local Binary Patterns for Land-Use Scene Classification

Gabor-Filtering-Based Completed Local Binary Patterns for Land-Use Scene Classification Gabor-Flterng-Based Completed Loal Bnary Patterns for Land-Use Sene Classfaton Chen Chen 1, Lbng Zhou 2,*, Janzhong Guo 1,2, We L 3, Hongjun Su 4, Fangda Guo 5 1 Department of Eletral Engneerng, Unversty

More information

TAR based shape features in unconstrained handwritten digit recognition

TAR based shape features in unconstrained handwritten digit recognition TAR based shape features n unonstraned handwrtten dgt reognton P. AHAMED AND YOUSEF AL-OHALI Department of Computer Sene Kng Saud Unversty P.O.B. 578, Ryadh 543 SAUDI ARABIA shamapervez@gmal.om, yousef@s.edu.sa

More information

Connectivity in Fuzzy Soft graph and its Complement

Connectivity in Fuzzy Soft graph and its Complement IOSR Journal of Mathemats (IOSR-JM) e-issn: 2278-5728, p-issn: 2319-765X. Volume 1 Issue 5 Ver. IV (Sep. - Ot.2016), PP 95-99 www.osrjournals.org Connetvty n Fuzzy Soft graph and ts Complement Shashkala

More information

Collaboratively Regularized Nearest Points for Set Based Recognition

Collaboratively Regularized Nearest Points for Set Based Recognition Academc Center for Computng and Meda Studes, Kyoto Unversty Collaboratvely Regularzed Nearest Ponts for Set Based Recognton Yang Wu, Mchhko Mnoh, Masayuk Mukunok Kyoto Unversty 9/1/013 BMVC 013 @ Brstol,

More information

Steganalysis of DCT-Embedding Based Adaptive Steganography and YASS

Steganalysis of DCT-Embedding Based Adaptive Steganography and YASS Steganalyss of DCT-Embeddng Based Adaptve Steganography and YASS Qngzhong Lu Department of Computer Sene Sam Houston State Unversty Huntsvlle, TX 77341, U.S.A. lu@shsu.edu ABSTRACT Reently well-desgned

More information

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Learning the Kernel Parameters in Kernel Minimum Distance Classifier Learnng the Kernel Parameters n Kernel Mnmum Dstance Classfer Daoqang Zhang 1,, Songcan Chen and Zh-Hua Zhou 1* 1 Natonal Laboratory for Novel Software Technology Nanjng Unversty, Nanjng 193, Chna Department

More information

Discriminative Dictionary Learning with Pairwise Constraints

Discriminative Dictionary Learning with Pairwise Constraints Dscrmnatve Dctonary Learnng wth Parwse Constrants Humn Guo Zhuoln Jang LARRY S. DAVIS UNIVERSITY OF MARYLAND Nov. 6 th, Outlne Introducton/motvaton Dctonary Learnng Dscrmnatve Dctonary Learnng wth Parwse

More information

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1 4/14/011 Outlne Dscrmnatve classfers for mage recognton Wednesday, Aprl 13 Krsten Grauman UT-Austn Last tme: wndow-based generc obect detecton basc ppelne face detecton wth boostng as case study Today:

More information

Improving Web Image Search using Meta Re-rankers

Improving Web Image Search using Meta Re-rankers VOLUME-1, ISSUE-V (Aug-Sep 2013) IS NOW AVAILABLE AT: www.dcst.com Improvng Web Image Search usng Meta Re-rankers B.Kavtha 1, N. Suata 2 1 Department of Computer Scence and Engneerng, Chtanya Bharath Insttute

More information

Lecture 5: Multilayer Perceptrons

Lecture 5: Multilayer Perceptrons Lecture 5: Multlayer Perceptrons Roger Grosse 1 Introducton So far, we ve only talked about lnear models: lnear regresson and lnear bnary classfers. We noted that there are functons that can t be represented

More information

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 1. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 1. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY SSDH: Sem-supervsed Deep Hashng for Large Scale Image Retreval Jan Zhang, and Yuxn Peng arxv:607.08477v2 [cs.cv] 8 Jun 207 Abstract Hashng

More information

Bottom-Up Fuzzy Partitioning in Fuzzy Decision Trees

Bottom-Up Fuzzy Partitioning in Fuzzy Decision Trees Bottom-Up Fuzzy arttonng n Fuzzy eson Trees Maej Fajfer ept. of Mathemats and Computer Sene Unversty of Mssour St. Lous St. Lous, Mssour 63121 maejf@me.pl Cezary Z. Janow ept. of Mathemats and Computer

More information

Proceedings of the International MultiConference of Engineers and Computer Scientists 2018 Vol I IMECS 2018, March 14-16, 2018, Hong Kong

Proceedings of the International MultiConference of Engineers and Computer Scientists 2018 Vol I IMECS 2018, March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong TABLE I CLASSIFICATION ACCURACY OF DIFFERENT PRE-TRAINED MODELS ON THE TEST DATA

More information

Integrating Fuzzy c-means Clustering with PostgreSQL *

Integrating Fuzzy c-means Clustering with PostgreSQL * SQL alle pgfcm. Seton 0 brefly susses relate work. Seton 0 ontans onluson remarks an retons for future work. Integratng Fuzzy -Means Clusterng wth PostgreSQL * R. M. Mnakhmetov taven@gmal.om South Ural

More information

Fuzzy Modeling for Multi-Label Text Classification Supported by Classification Algorithms

Fuzzy Modeling for Multi-Label Text Classification Supported by Classification Algorithms Journal of Computer Senes Orgnal Researh Paper Fuzzy Modelng for Mult-Label Text Classfaton Supported by Classfaton Algorthms 1 Beatrz Wlges, 2 Gustavo Mateus, 2 Slva Nassar, 2 Renato Cslagh and 3 Rogéro

More information

A Fast Way to Produce Optimal Fixed-Depth Decision Trees

A Fast Way to Produce Optimal Fixed-Depth Decision Trees A Fast Way to Produe Optmal Fxed-Depth Deson Trees Alreza Farhangfar, Russell Grener and Martn Znkevh Dept of Computng Sene Unversty of Alberta Edmonton, Alberta T6G 2E8 Canada {farhang, grener, maz}@s.ualberta.a

More information

Pattern Classification: An Improvement Using Combination of VQ and PCA Based Techniques

Pattern Classification: An Improvement Using Combination of VQ and PCA Based Techniques Ameran Journal of Appled Senes (0): 445-455, 005 ISSN 546-939 005 Sene Publatons Pattern Classfaton: An Improvement Usng Combnaton of VQ and PCA Based Tehnques Alok Sharma, Kuldp K. Palwal and Godfrey

More information

Cluster ( Vehicle Example. Cluster analysis ( Terminology. Vehicle Clusters. Why cluster?

Cluster (  Vehicle Example. Cluster analysis (  Terminology. Vehicle Clusters. Why cluster? Why luster? referene funton R R Although R and R both somewhat orrelated wth the referene funton, they are unorrelated wth eah other Cluster (www.m-w.om) A number of smlar ndvduals that our together as

More information

A Real-Time Detecting Algorithm for Tracking Community Structure of Dynamic Networks

A Real-Time Detecting Algorithm for Tracking Community Structure of Dynamic Networks A Real-Tme Detetng Algorthm for Trakng Communty Struture of Dynam Networks Jaxng Shang*, Lanhen Lu*, Feng Xe, Zhen Chen, Jaa Mao, Xueln Fang, Cheng Wu* Department of Automaton, Tsnghua Unversty, Beng,,

More information

Machine Learning 9. week

Machine Learning 9. week Machne Learnng 9. week Mappng Concept Radal Bass Functons (RBF) RBF Networks 1 Mappng It s probably the best scenaro for the classfcaton of two dataset s to separate them lnearly. As you see n the below

More information

Pairwise Identity Verification via Linear Concentrative Metric Learning

Pairwise Identity Verification via Linear Concentrative Metric Learning Parwse Identty Verfaton va Lnear Conentratve Metr Learnng Lle Zheng, Stefan Duffner, Khald Idrss, Chrstophe Gara, Atlla Baskurt To te ths verson: Lle Zheng, Stefan Duffner, Khald Idrss, Chrstophe Gara,

More information

Bit-level Arithmetic Optimization for Carry-Save Additions

Bit-level Arithmetic Optimization for Carry-Save Additions Bt-leel Arthmet Optmzaton for Carry-Sae s Ke-Yong Khoo, Zhan Yu and Alan N. Wllson, Jr. Integrated Cruts and Systems Laboratory Unersty of Calforna, Los Angeles, CA 995 khoo, zhanyu, wllson @sl.ula.edu

More information

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS P.G. Demdov Yaroslavl State Unversty Anatoly Ntn, Vladmr Khryashchev, Olga Stepanova, Igor Kostern EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS Yaroslavl, 2015 Eye

More information

Active Contours/Snakes

Active Contours/Snakes Actve Contours/Snakes Erkut Erdem Acknowledgement: The sldes are adapted from the sldes prepared by K. Grauman of Unversty of Texas at Austn Fttng: Edges vs. boundares Edges useful sgnal to ndcate occludng

More information

Performance Analysis of Hybrid (supervised and unsupervised) method for multiclass data set

Performance Analysis of Hybrid (supervised and unsupervised) method for multiclass data set IOSR Journal of Computer Engneerng (IOSR-JCE) e-issn: 2278-0661,p-ISSN: 2278-8727, Volume 16, Issue 4, Ver. III (Jul Aug. 2014), PP 93-99 www.osrjournals.org Performane Analyss of Hybrd (supervsed and

More information

Fuzzy C-Means Initialized by Fixed Threshold Clustering for Improving Image Retrieval

Fuzzy C-Means Initialized by Fixed Threshold Clustering for Improving Image Retrieval Fuzzy -Means Intalzed by Fxed Threshold lusterng for Improvng Image Retreval NAWARA HANSIRI, SIRIPORN SUPRATID,HOM KIMPAN 3 Faculty of Informaton Technology Rangst Unversty Muang-Ake, Paholyotn Road, Patumtan,

More information

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data A Fast Content-Based Multmeda Retreval Technque Usng Compressed Data Borko Furht and Pornvt Saksobhavvat NSF Multmeda Laboratory Florda Atlantc Unversty, Boca Raton, Florda 3343 ABSTRACT In ths paper,

More information

Fusion of Deep Features and Weighted VLAD Vectors based on Multiple Features for Image Retrieval

Fusion of Deep Features and Weighted VLAD Vectors based on Multiple Features for Image Retrieval MATEC Web of Conferences, 0500 (07) DTS-07 DO: 005/matecconf/070500 Fuson of Deep Features and Weghted VLAD Vectors based on Multple Features for mage Retreval Yanhong Wang,, Ygang Cen,, Lequan Lang,*,

More information

CS246: Mining Massive Datasets Jure Leskovec, Stanford University

CS246: Mining Massive Datasets Jure Leskovec, Stanford University CS46: Mnng Massve Datasets Jure Leskovec, Stanford Unversty http://cs46.stanford.edu /19/013 Jure Leskovec, Stanford CS46: Mnng Massve Datasets, http://cs46.stanford.edu Perceptron: y = sgn( x Ho to fnd

More information

Avatar Face Recognition using Wavelet Transform and Hierarchical Multi-scale LBP

Avatar Face Recognition using Wavelet Transform and Hierarchical Multi-scale LBP 2011 10th Internatonal Conferene on Mahne Learnng and Applatons Avatar Fae Reognton usng Wavelet Transform and Herarhal Mult-sale LBP Abdallah A. Mohamed, Darryl D Souza, Naouel Bal and Roman V. Yampolsky

More information

International Journal of Pharma and Bio Sciences HYBRID CLUSTERING ALGORITHM USING POSSIBILISTIC ROUGH C-MEANS ABSTRACT

International Journal of Pharma and Bio Sciences HYBRID CLUSTERING ALGORITHM USING POSSIBILISTIC ROUGH C-MEANS ABSTRACT Int J Pharm Bo S 205 Ot; 6(4): (B) 799-80 Researh Artle Botehnology Internatonal Journal of Pharma and Bo Senes ISSN 0975-6299 HYBRID CLUSTERING ALGORITHM USING POSSIBILISTIC ROUGH C-MEANS *ANURADHA J,

More information

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification Introducton to Artfcal Intellgence V22.0472-001 Fall 2009 Lecture 24: Nearest-Neghbors & Support Vector Machnes Rob Fergus Dept of Computer Scence, Courant Insttute, NYU Sldes from Danel Yeung, John DeNero

More information

Joint Object and Part Segmentation using Deep Learned Potentials

Joint Object and Part Segmentation using Deep Learned Potentials Jont Object and Part Segmentaton usng Deep Learned Potentals Peng Wang 1 Xaohu Shen 2 Zhe Ln 2 Scott Cohen 2 Bran Prce 2 Alan Yulle 1 1 Unversty of Calforna, Los Angeles 2 Adobe Research Abstract Segmentng

More information

Pixel-Based Texture Classification of Tissues in Computed Tomography

Pixel-Based Texture Classification of Tissues in Computed Tomography Pxel-Based Texture Classfaton of Tssues n Computed Tomography Ruhaneewan Susomboon, Danela Stan Rau, Jaob Furst Intellgent ultmeda Proessng Laboratory Shool of Computer Sene, Teleommunatons, and Informaton

More information

Determining the Optimal Bandwidth Based on Multi-criterion Fusion

Determining the Optimal Bandwidth Based on Multi-criterion Fusion Proceedngs of 01 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 5 (01) (01) IACSIT Press, Sngapore Determnng the Optmal Bandwdth Based on Mult-crteron Fuson Ha-L Lang 1+, Xan-Mn

More information

Evaluation of Segmentation in Magnetic Resonance Images Using k-means and Fuzzy c-means Clustering Algorithms

Evaluation of Segmentation in Magnetic Resonance Images Using k-means and Fuzzy c-means Clustering Algorithms ELEKTROTEHIŠKI VESTIK 79(3): 129-134, 2011 EGLISH EDITIO Evaluaton of Segmentaton n Magnet Resonane Images Usng k-means and Fuzzy -Means Clusterng Algorthms Tomaž Fnkšt Unverza v Lublan, Fakulteta za stronštvo,

More information

Fitting: Deformable contours April 26 th, 2018

Fitting: Deformable contours April 26 th, 2018 4/6/08 Fttng: Deformable contours Aprl 6 th, 08 Yong Jae Lee UC Davs Recap so far: Groupng and Fttng Goal: move from array of pxel values (or flter outputs) to a collecton of regons, objects, and shapes.

More information

arxiv: v2 [cs.cv] 9 Apr 2018

arxiv: v2 [cs.cv] 9 Apr 2018 Boundary-senstve Network for Portrat Segmentaton Xanzh Du 1, Xaolong Wang 2, Dawe L 2, Jngwen Zhu 2, Serafettn Tasc 2, Cameron Uprght 2, Stephen Walsh 2, Larry Davs 1 1 Computer Vson Lab, UMIACS, Unversty

More information

ALEXNET FEATURE EXTRACTION AND MULTI-KERNEL LEARNING FOR OBJECT- ORIENTED CLASSIFICATION

ALEXNET FEATURE EXTRACTION AND MULTI-KERNEL LEARNING FOR OBJECT- ORIENTED CLASSIFICATION ALEXNET FEATURE EXTRACTION AND MULTI-KERNEL LEARNING FOR OBJECT- ORIENTED CLASSIFICATION Lng Dng 1, Hongy L 2, *, Changmao Hu 2, We Zhang 2, Shumn Wang 1 1 Insttute of Earthquake Forecastng, Chna Earthquake

More information

Deep learning is a good steganalysis tool when embedding key is reused for different images, even if there is a cover source-mismatch

Deep learning is a good steganalysis tool when embedding key is reused for different images, even if there is a cover source-mismatch Deep learnng s a good steganalyss tool when embeddng key s reused for dfferent mages, even f there s a cover source-msmatch Lonel PIBRE 2,3, Jérôme PASQUET 2,3, Dno IENCO 2,3, Marc CHAUMONT 1,2,3 (1) Unversty

More information

Image Alignment CSC 767

Image Alignment CSC 767 Image Algnment CSC 767 Image algnment Image from http://graphcs.cs.cmu.edu/courses/15-463/2010_fall/ Image algnment: Applcatons Panorama sttchng Image algnment: Applcatons Recognton of object nstances

More information

Progressive scan conversion based on edge-dependent interpolation using fuzzy logic

Progressive scan conversion based on edge-dependent interpolation using fuzzy logic Progressve san onverson based on edge-dependent nterpolaton usng fuzzy log P. Brox brox@mse.nm.es I. Baturone lum@mse.nm.es Insttuto de Mroeletróna de Sevlla, Centro Naonal de Mroeletróna Avda. Rena Meredes

More information

Smoothing Spline ANOVA for variable screening

Smoothing Spline ANOVA for variable screening Smoothng Splne ANOVA for varable screenng a useful tool for metamodels tranng and mult-objectve optmzaton L. Rcco, E. Rgon, A. Turco Outlne RSM Introducton Possble couplng Test case MOO MOO wth Game Theory

More information

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching A Fast Vsual Trackng Algorthm Based on Crcle Pxels Matchng Zhqang Hou hou_zhq@sohu.com Chongzhao Han czhan@mal.xjtu.edu.cn Ln Zheng Abstract: A fast vsual trackng algorthm based on crcle pxels matchng

More information

AVideoStabilizationMethodbasedonInterFrameImageMatchingScore

AVideoStabilizationMethodbasedonInterFrameImageMatchingScore Global Journal of Computer Sene and Tehnology: F Graphs & vson Volume 7 Issue Verson.0 Year 207 Type: Double Blnd Peer Revewed Internatonal Researh Journal Publsher: Global Journals In. (USA) Onlne ISSN:

More information

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices Steps for Computng the Dssmlarty, Entropy, Herfndahl-Hrschman and Accessblty (Gravty wth Competton) Indces I. Dssmlarty Index Measurement: The followng formula can be used to measure the evenness between

More information

Switching Convolutional Neural Network for Crowd Counting

Switching Convolutional Neural Network for Crowd Counting Swtchng Convolutonal Neural Network for Crowd Countng Deepak Babu Sam Shv Surya R. Venkatesh Babu Indan Insttute of Scence Bangalore, INDIA 560012 bsdeepak@grads.cds.sc.ac.n, shv.surya314@gmal.com, venky@cds.sc.ac.n

More information

An Image Fusion Approach Based on Segmentation Region

An Image Fusion Approach Based on Segmentation Region Rong Wang, L-Qun Gao, Shu Yang, Yu-Hua Cha, and Yan-Chun Lu An Image Fuson Approach Based On Segmentaton Regon An Image Fuson Approach Based on Segmentaton Regon Rong Wang, L-Qun Gao, Shu Yang 3, Yu-Hua

More information

Learning Convention Propagation in BeerAdvocate Reviews from a etwork Perspective. Abstract

Learning Convention Propagation in BeerAdvocate Reviews from a etwork Perspective. Abstract CS 9 Projet Final Report: Learning Convention Propagation in BeerAdvoate Reviews from a etwork Perspetive Abstrat We look at the way onventions propagate between reviews on the BeerAdvoate dataset, and

More information

Support Vector Machines. CS534 - Machine Learning

Support Vector Machines. CS534 - Machine Learning Support Vector Machnes CS534 - Machne Learnng Perceptron Revsted: Lnear Separators Bnar classfcaton can be veed as the task of separatng classes n feature space: b > 0 b 0 b < 0 f() sgn( b) Lnear Separators

More information

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET

BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET 1 BOOSTING CLASSIFICATION ACCURACY WITH SAMPLES CHOSEN FROM A VALIDATION SET TZU-CHENG CHUANG School of Electrcal and Computer Engneerng, Purdue Unversty, West Lafayette, Indana 47907 SAUL B. GELFAND School

More information

A Robust Algorithm for Text Detection in Color Images

A Robust Algorithm for Text Detection in Color Images A Robust Algorthm for Tet Deteton n Color Images Yangng LIU Satosh GOTO Takesh IKENAGA Abstrat Tet deteton n olor mages has beome an atve researh area sne reent deades. In ths paper we present a novel

More information

SEMANTIC image segmentation has been one of the most

SEMANTIC image segmentation has been one of the most 1 Stacked Deconvolutonal Network for Semantc Segmentaton Jun Fu, Jng Lu, Member, IEEE, Yuhang Wang, and Hanqng Lu, Senor Member, IEEE arxv:1708.04943v1 [cs.cv] 16 Aug 2017 Abstract Recent progress n semantc

More information

Interval uncertain optimization of structures using Chebyshev meta-models

Interval uncertain optimization of structures using Chebyshev meta-models 0 th World Congress on Strutural and Multdsplnary Optmzaton May 9-24, 203, Orlando, Florda, USA Interval unertan optmzaton of strutures usng Chebyshev meta-models Jngla Wu, Zhen Luo, Nong Zhang (Tmes New

More information

Joint Example-based Depth Map Super-Resolution

Joint Example-based Depth Map Super-Resolution Jont Example-based Depth Map Super-Resoluton Yanje L 1, Tanfan Xue,3, Lfeng Sun 1, Janzhuang Lu,3,4 1 Informaton Scence and Technology Department, Tsnghua Unversty, Bejng, Chna Department of Informaton

More information

Research of Image Recognition Algorithm Based on Depth Learning

Research of Image Recognition Algorithm Based on Depth Learning 208 4th World Conference on Control, Electroncs and Computer Engneerng (WCCECE 208) Research of Image Recognton Algorthm Based on Depth Learnng Zhang Jan, J Xnhao Zhejang Busness College, Hangzhou, Chna,

More information

arxiv: v2 [cs.cv] 3 Aug 2017

arxiv: v2 [cs.cv] 3 Aug 2017 Swtchng Convolutonal Neural Network for Crowd Countng Deepak Babu Sam Shv Surya R. Venkatesh Babu Indan Insttute of Scence Bangalore, INDIA 560012 arxv:1708.00199v2 [cs.cv] 3 Aug 2017 bsdeepak@grads.cds.sc.ac.n,

More information

arxiv: v1 [cs.cv] 4 Jul 2016

arxiv: v1 [cs.cv] 4 Jul 2016 JUN et al.: LEARNING FINE-SCALED DEPTH MAPS FROM SINGLE RGB IMAGES 1 arxv:1607.00730v1 [cs.cv] 4 Jul 2016 Learnng Fne-Scaled Depth Maps from Sngle RGB Images Jun L http://cg.cs.un-bonn.de/ Renhard Klen

More information

Feature-Fused SSD: Fast Detection for Small Objects

Feature-Fused SSD: Fast Detection for Small Objects Feature-Fused SSD: Fast Detection for Small Objects Guimei Cao, Xuemei Xie, Wenzhe Yang, Quan Liao, Guangming Shi, Jinjian Wu School of Electronic Engineering, Xidian University, China xmxie@mail.xidian.edu.cn

More information

Backpropagation: In Search of Performance Parameters

Backpropagation: In Search of Performance Parameters Bacpropagaton: In Search of Performance Parameters ANIL KUMAR ENUMULAPALLY, LINGGUO BU, and KHOSROW KAIKHAH, Ph.D. Computer Scence Department Texas State Unversty-San Marcos San Marcos, TX-78666 USA ae049@txstate.edu,

More information

FUZZY SEGMENTATION IN IMAGE PROCESSING

FUZZY SEGMENTATION IN IMAGE PROCESSING FUZZY SEGMENTATION IN IMAGE PROESSING uevas J. Er,, Zaldívar N. Danel,, Roas Raúl Free Unverstät Berln, Insttut für Inforat Tausstr. 9, D-495 Berln, Gerany. Tel. 0049-030-8385485, Fax. 0049-030-8387509

More information

Tone-Aware Sparse Representation for Face Recognition

Tone-Aware Sparse Representation for Face Recognition Tone-Aware Sparse Representaton for Face Recognton Lngfeng Wang, Huayu Wu and Chunhong Pan Abstract It s stll a very challengng task to recognze a face n a real world scenaro, snce the face may be corrupted

More information

Deep learning for object detection. Slides from Svetlana Lazebnik and many others

Deep learning for object detection. Slides from Svetlana Lazebnik and many others Deep learning for object detection Slides from Svetlana Lazebnik and many others Recent developments in object detection 80% PASCAL VOC mean0average0precision0(map) 70% 60% 50% 40% 30% 20% 10% Before deep

More information

Robust Dictionary Learning with Capped l 1 -Norm

Robust Dictionary Learning with Capped l 1 -Norm Proceedngs of the Twenty-Fourth Internatonal Jont Conference on Artfcal Intellgence (IJCAI 205) Robust Dctonary Learnng wth Capped l -Norm Wenhao Jang, Fepng Ne, Heng Huang Unversty of Texas at Arlngton

More information

Optimal shape and location of piezoelectric materials for topology optimization of flextensional actuators

Optimal shape and location of piezoelectric materials for topology optimization of flextensional actuators Optmal shape and loaton of pezoeletr materals for topology optmzaton of flextensonal atuators ng L 1 Xueme Xn 2 Noboru Kkuh 1 Kazuhro Satou 1 1 Department of Mehanal Engneerng, Unversty of Mhgan, Ann Arbor,

More information

Load Balancing for Hex-Cell Interconnection Network

Load Balancing for Hex-Cell Interconnection Network Int. J. Communcatons, Network and System Scences,,, - Publshed Onlne Aprl n ScRes. http://www.scrp.org/journal/jcns http://dx.do.org/./jcns.. Load Balancng for Hex-Cell Interconnecton Network Saher Manaseer,

More information

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 25, NO. 4, APRIL

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 25, NO. 4, APRIL IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 25, NO. 4, APRIL 2016 1713 Weakly Supervsed Fne-Graned Categorzaton Wth Part-Based Image Representaton Yu Zhang, Xu-Shen We, Janxn Wu, Member, IEEE, Janfe Ca,

More information

Minimize Congestion for Random-Walks in Networks via Local Adaptive Congestion Control

Minimize Congestion for Random-Walks in Networks via Local Adaptive Congestion Control Journal of Communatons Vol. 11, No. 6, June 2016 Mnmze Congeston for Random-Walks n Networks va Loal Adaptve Congeston Control Yang Lu, Y Shen, and Le Dng College of Informaton Sene and Tehnology, Nanjng

More information

An Optimal Algorithm for Prufer Codes *

An Optimal Algorithm for Prufer Codes * J. Software Engneerng & Applcatons, 2009, 2: 111-115 do:10.4236/jsea.2009.22016 Publshed Onlne July 2009 (www.scrp.org/journal/jsea) An Optmal Algorthm for Prufer Codes * Xaodong Wang 1, 2, Le Wang 3,

More information

Deep Spatial-Temporal Joint Feature Representation for Video Object Detection

Deep Spatial-Temporal Joint Feature Representation for Video Object Detection sensors Artcle Deep Spatal-Temporal Jont Feature Representaton for Vdeo Object Detecton Baojun Zhao 1,2, Boya Zhao 1,2 ID, Lnbo Tang 1,2, *, Yuq Han 1,2 and Wenzheng Wang 1,2 1 School of Informaton and

More information

Face Detection with Deep Learning

Face Detection with Deep Learning Face Detecton wth Deep Learnng Yu Shen Yus122@ucsd.edu A13227146 Kuan-We Chen kuc010@ucsd.edu A99045121 Yzhou Hao y3hao@ucsd.edu A98017773 Mn Hsuan Wu mhwu@ucsd.edu A92424998 Abstract The project here

More information

GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation

GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation GeoNet: Geometrc Neural Network for Jont Depth and Surface Normal Estmaton Xaojuan Q Renje Lao, Zhengzhe Lu Raquel Urtasun, Jaya Ja, The Chnese Unversty of Hong Kong Unversty of Toronto Uber Advanced Technologes

More information

Bilateral Mesh Denoising

Bilateral Mesh Denoising Outlne Blateral Meh Denong S. Flehman, I. Dror,, D. Cohen-Or Tel Avv Unverty Preented by Derek Bradley Motvaton Prevou ork Blateral Meh Denong Image Proeng Bakground Blateral Image Flterng Tranformng from

More information

Cluster Analysis of Electrical Behavior

Cluster Analysis of Electrical Behavior Journal of Computer and Communcatons, 205, 3, 88-93 Publshed Onlne May 205 n ScRes. http://www.scrp.org/ournal/cc http://dx.do.org/0.4236/cc.205.350 Cluster Analyss of Electrcal Behavor Ln Lu Ln Lu, School

More information

ON THE USE OF THE SIFT TRANSFORM TO SELF-LOCATE AND POSITION EYE-IN-HAND MANIPULATORS USING VISUAL CONTROL

ON THE USE OF THE SIFT TRANSFORM TO SELF-LOCATE AND POSITION EYE-IN-HAND MANIPULATORS USING VISUAL CONTROL XVIII Congresso Braslero de Automáta / a 6-setembro-00, Bonto-MS ON THE USE OF THE SIFT TRANSFORM TO SELF-LOCATE AND POSITION EYE-IN-HAND MANIPULATORS USING VISUAL CONTROL ILANA NIGRI, RAUL Q. FEITOSA

More information

Available online at Available online at Advanced in Control Engineering and Information Science

Available online at   Available online at   Advanced in Control Engineering and Information Science Avalable onlne at wwwscencedrectcom Avalable onlne at wwwscencedrectcom Proceda Proceda Engneerng Engneerng 00 (2011) 15000 000 (2011) 1642 1646 Proceda Engneerng wwwelsevercom/locate/proceda Advanced

More information

Deep-dense Conditional Random Fields for Object Co-segmentation

Deep-dense Conditional Random Fields for Object Co-segmentation Deep-dense Condtonal Random Felds for Object Co-segmentaton Zehuan Yuan 1, Tong Lu 1, and Yru Wu 2 1 State Key Laboratory for Novel Software Technology, Nanjng Unversty, Chna 2 College of Computer and

More information

Kernel Collaborative Representation Classification Based on Adaptive Dictionary Learning

Kernel Collaborative Representation Classification Based on Adaptive Dictionary Learning Internatonal Journal of Intellgent Informaton Systems 2018; 7(2): 15-22 http://www.scencepublshnggroup.com/j/js do: 10.11648/j.js.20180702.11 ISSN: 2328-7675 (Prnt); ISSN: 2328-7683 (Onlne) Kernel Collaboratve

More information

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour 6.854 Advanced Algorthms Petar Maymounkov Problem Set 11 (November 23, 2005) Wth: Benjamn Rossman, Oren Wemann, and Pouya Kheradpour Problem 1. We reduce vertex cover to MAX-SAT wth weghts, such that the

More information

Clustering incomplete data using kernel-based fuzzy c-means algorithm

Clustering incomplete data using kernel-based fuzzy c-means algorithm Clusterng noplete data usng ernel-based fuzzy -eans algorth Dao-Qang Zhang *, Song-Can Chen Departent of Coputer Sene and Engneerng, Nanjng Unversty of Aeronauts and Astronauts, Nanjng, 210016, People

More information

,.,,

,.,, ISSN 49-99 6. 9... /.......... 989.... 85-9.... - /.... //.. 5.. 8.. 5-55. 4... /.... //... 978.... 65-7. 5... :.... - :.: 5..6. /. 99. 46. Vtor su leturer Vtor alh Prof. PhD teh. s. lexandr Ddy ssos.

More information

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning

Outline. Type of Machine Learning. Examples of Application. Unsupervised Learning Outlne Artfcal Intellgence and ts applcatons Lecture 8 Unsupervsed Learnng Professor Danel Yeung danyeung@eee.org Dr. Patrck Chan patrckchan@eee.org South Chna Unversty of Technology, Chna Introducton

More information

arxiv: v1 [cs.cv] 23 Apr 2015

arxiv: v1 [cs.cv] 23 Apr 2015 Person Re-dentfcaton wth Correspondence Structure Learnng Yang Shen 1, Weyao Ln 1, Junch Yan 1, Mnglang Xu 2, Janxn Wu 3 and Jngdong Wang 4 1 Shangha Jao Tong Unversty, Chna 2 Zhengzhou Unversty, Chna

More information

An Adaptive Filter Based on Wavelet Packet Decomposition in Motor Imagery Classification

An Adaptive Filter Based on Wavelet Packet Decomposition in Motor Imagery Classification An Adaptve Flter Based on Wavelet Paket Deomposton n Motor Imagery Classfaton J. Payat, R. Mt, T. Chusak, and N. Sugno Abstrat Bran-Computer Interfae (BCI) s a system that translates bran waves nto eletral

More information

Microprocessors and Microsystems

Microprocessors and Microsystems Mroproessors and Mrosystems 36 (2012) 96 109 Contents lsts avalable at SeneDret Mroproessors and Mrosystems journal homepage: www.elsever.om/loate/mpro Hardware aelerator arhteture for smultaneous short-read

More information

High-Boost Mesh Filtering for 3-D Shape Enhancement

High-Boost Mesh Filtering for 3-D Shape Enhancement Hgh-Boost Mesh Flterng for 3-D Shape Enhancement Hrokazu Yagou Λ Alexander Belyaev y Damng We z Λ y z ; ; Shape Modelng Laboratory, Unversty of Azu, Azu-Wakamatsu 965-8580 Japan y Computer Graphcs Group,

More information

Odor Recognition in Multiple E-nose Systems with Cross-domain Discriminative Subspace Learning

Odor Recognition in Multiple E-nose Systems with Cross-domain Discriminative Subspace Learning hs paper has been aepted for publaton n IEEE ransatons on Instrumentaton and easurement, 7 Odor Reognton n ultple E-nose ystems wth Cross-doman Dsrmnatve ubspae Learnng Le Zhang, ember, IEEE, Yan Lu, and

More information