Real-Time Multi-View Face Detection
|
|
- Erick Wells
- 6 years ago
- Views:
Transcription
1 Real-Time Multi-View Face Detectin ZhenQiu Zhang 1*, Lng Zhu 2, Stan Z. Li 2, HngJiang Zhang 2 1. Institute f Autmatin, Chinese Academy f Science, Beijing, China 2. Micrsft Research Asia, Beijing Sigma Center, Beijing , China Abstract In this paper, we present a detectr-pyramid architecture fr real-time multi-view face detectin. Using a carse t fine strategy, the full view is partitined int finer and finer views. Each face detectr in the pyramid detects faces f its respective view range. Its training is perfrmed by using a new meta bting learning algrithm. This results in the first real-time multi-view face detectin system which runs at 5 frames per secnd fr 320x240 image sequence. 1. Intrductin Statistics shw that apprximately 75% f the faces in hme phts are nn-frntal [1], and therefre ability t deal with multi-view faces is imprtant fr many face-related applicatins. Multi-view face detectin has been a challenging prblem. The challenge is firstly due t large amunt f variatin and cmplexity brught abut by the changes in facial appearance, lighting and expressin [17]. Changes in facial view (pse) further cmplicate the situatin because the distributin f multi-view faces in a feature space is mre dispersed and mre cmplicated than that f frntal faces. The learning based apprach has been mst effective fr face detectin. Sung and Pggi [14] divided the frntal face image space and nn-face image space each int several prbability clusters. PCA is perfrmed n each cluster s that face/nn-face classificatin is perfrmed in terms f bth the Mahalanbis distance frm the cluster center in the PCA space and the Euclidean distance frm the PCA space are used as features. Rwley et al [9] presented a face detectin system based n retinally cnnected neural netwrks. The input t the NN is the preprcessed image pixel values directly. Pst-prcessing f the neural netwrks are perfrmed by either ANDing/ORing the utputs r using an additinal neural netwrk t arbitrate between the utputs. Osuna el al [5] applied the supprt vectr machines algrithm t train an NN t classify face and nn-face patterns. Yang el al [3] uses a netwrk f linear units. The SNW learning architecture is specifically tailred fr learning in the presence f a very large number f features. * The wrk presented in the paper was carried ut at Micrsft Research Asia. Recently, Vila and Jnes [19] prpse a very fast apprach fr frntal face detectin. Simple Haar-like feature are extracted, face/nn-face classificatin is dne by using a cascade f successively mre cmplex classifiers which are trained by using AdaBst [24] learning algrithm. The cascade structure is supprted by an argument made in [4] that cascading classifiers is a better apprach than multiexpert methds like vting and stacking. Over past years, prgress has been made fr nn-frntal faces detectin and recgnitin. Feraud et al [11] adpt the view-based representatin fr face detectin, and use an array f 5 detectrs with each detectr respnsible fr ne view. Wisktt et al [16] build elastic bunch graph templates fr multi-view face detectin and recgnitin. Gng and clleagues [21] study the trajectries f faces in linear PCA feature spaces as they rtate, and use kernel supprt vectr machines (SVMs) fr multi-pse face detectin and pse estimatin. Huang et al [13] use an SVM t classify three facial pses at , 0, degrees. T deal with cmplexity due t multi-view, a natural treatment is t divide face images int several subsets accrding t the facial view and mdel each view subspace respectively [2], by which explicit 3D mdeling is avided. The system f Schneiderman and Kanade is claimed t be the first ne in the wrld fr multi-view face detectin [10]. The algrithm cnsists f an array f 5 face detectrs each f which is specialized fr a specific pse f face and accmmdates small amunt f variatin arund the designated pse. A detectr classifies a sub-windw int face/nn-face based n statistics f prducts f histgrams cmputed frm examples f the respective view. The results frm all the detectrs are merged such that they are spatially cnsistent. The detectr is claimed t be the first algrithm in the wrld fr multi-view face detectin. Hwever, it is very slw and takes 1 min t wrk n a 320x240 image ver nly 4 ctaves f candidate size [10]. In this paper, we present a nvel framewrk fr real-time multi-view face detectin. A detectr-pyramid architecture is designed t detect multi-view faces efficiently. The detectrpyramid adpts an integrated strategy f carse-t-fine view decmpsitin [18,19], and simple-t-cmplex face/nnface classificatin Vila and Jnes [19]; a sub-windw is prcessed frm the tp t bttm f the pyramid by a sequence f increasingly mre cmplex face/nn-face classifiers designed fr increasingly finer ranges f facial view. The detectr-pyramid ges beynd the straightfrward view decmpsitin methd [2] in that using the carse-tfine and simple-t-cmplex strategy, a vast number f
2 nnface sub-windws can be discarded very quickly with very little lse f face sub-windws. This is very imprtant fr fast face detectin because nly a tiny prprtin f subwindws are f faces. We devise simple image features fr efficient face/nnface classificatin. These features are extensins f thse used in [19] fr frntal face detectin in that the frmer is mre suitable t cater t nn-symmetry f nnfrntal faces. Every detectr in the pyramid is learned frm face/nnface examples using a new learning algrithm called FlatBst [22]. FlatBst incrprates the idea f Flating Search [18] int AdaBst t slve the nnmntnicity prblem encuntered in the sequential algrithm f AdaBst. While the Vila-Jnes detectr [19] is the first real-time frntal face detectr and Schneiderman-Kanade detectr is the first (nn real-time) multi-view face detectr, the algrithm presented in this paper results in the first real-time multi-view face detectin system which runs at 5 frames per secnd fr 320x240 image sequence n a cnventinal 700 MHz Pentium III PC. The rest f the paper is rganized as fllws: Sectin 2 intrduces the detectr-pyramid architecture fr multi-view face detectin. The design and training f individual detectr are presented in sectin 3 and 4. Methd t arbitrate amng nine view channels is presented in sectin 5.Sectin 6 prvides the experimental results and cnclusin is drawn in sectin Detectr-Pyramid Architecture The present multi-view face detectin system is distinguished frm previus systems in its ability t detect multi-view faces in real-time. It is designed based n the fllwing thughts: While it is extremely difficult t distinguish multi-view faces frm nn-face images clearly using a single classifier, it is less difficult t classify between frntal faces and nn-faces and als less difficult t d between multi-view faces and part f nn-faces. Therefre, narrwing dwn the range f view will make face detectin easier and mre accurate fr that view. On the ther hand, a vast number f sub-windws (e.g. 70,401 square sub-windws can result frm the scan f a 320x240 image, frm the size f 20x20 pixels t 240x240 fr the size increment factr f 1.25) result frm scan f the input image; amng these nly a tiny prprtin (say, up t a few dzens) f them are faces. It can save the cmputatin tremendusly if a sequence f detectrs f increasing cmplexity and face/nn-face discriminating pwer are applied t quickly discard nn-faces at the earliest pssible stage using the simplest pssible features. The detectr-pyramid architecture (see Figure 1) is mtivated by the abve reasns. It adpts the carse t fine (tp-dwn in the pyramid) strategy [18,19] in that the full range f facial view is partitined int increasingly narrwer ranges, and thereby the whle face space is partitined int increasingly smaller subspaces. Als it adpts the simple-tcmplex strategy (Vila-Jnes detectr [19]) in that the earlier nes are simpler and s are able t reject a vast number f nn-face sub-windws q1uickly whereas the nes in the later stage are mre cmplex and invlved and spend mre time t scrutinize nly a relatively tiny number f remaining sub-windws. Figure 1: Detectr-pyramid. Our current implementatin cnsists f three levels. The first level cnsists f a single detectr, respnsible fr the full range f [-90,90] degree (0 degree being the frntal; view). There are three detectrs in the secnd level, respnsible fr the three view ranges [-90, -40], [-30,+30], [+40,+90], respectively. The third level cnsists f 9 detectrs f [-90, -80],[-70,-60],, [60,70], [80,90] degrees. Therefre, there are a ttal f 13 detectrs. Fr a sub-windw, if it is rejected by detectr at the tp level, it will be seen as nn-face regin and will nt be prcessed by later levels. If it ges thrugh first level, it will be prcessed by secnd level. If any detectr in secnd level classifies it as face, it will be prcessed by last level, r it will be rejected as nn-face. There are much mre detectrs n the bttm f ur framewrk, and it help us fcus ur attentin n thse pssible face regin, while paying much less time n impssible face regin. At the last level, each detectr nly dues with 20 degree ranges f view and each detectr has high detectin rate fr that view. This pyramidlike framewrk makes ur system have bth high detectin rate and rapid detectin speed fr multi-view face detectin. The full-view detectr in the implementatin is able t reject abut 50% f nn-face sub-windws scanned in the perfrming stage, while retaining 99% f training face examples in the training stage. Only retained sub-windws pssibly cntaining faces are further prcessed in the subsequent levels f finer detectrs. The results frm the detectrs in the bttm level are merged t make a final decisin regarding the input sub-windw. 3. Design f Individual Detectrs The high speed and detectin rate f the algrithm depend nt nly n the detectr-pyramid architecture but als individual detectr. Each detectr classifies a subwindw int face/nn-face. Tw types f simple features,
3 which are blck differences similar t steerable filters, are cmputed as shwn in Figure 2. Each such feature has a scalar value which can be cmputed very efficiently frm the summed-area table [6] r integral image [19]. These features are nn-symmetrical t cater t nnsymmetrical characteristics f nn-frntal faces. They have mre degrees f freedm than thse f [19] in their cnfiguratins: 6 (x, y, delta x, delta y, dx, dy) in the tw blck features and 7 (x, y, delta x, delta y, dx, dx, dy) in the three blck features. There are a ttal number f 102,979 tw-blck features fr a sub-windw f size 20x20 pixels. There are a ttal number f 188,366 three-blck features (with sme restrict t their freedm). x -2 dx y dy +2 Figure 2: The tw types f simple Harr wavelet like features defined in a sub-windw. The rectangles are f size x by y and are at distances f (dx,dy) apart. Each feature takes a value calculated by the weighted +-1,2 sum f the pixels in the rectangles. A face/nnface strng classifier is cnstructed based n a number f weak classifiers where a weak classifier perfrms face/nn-face classificatin using a different single feature, e.g. by threshlding the scalar value f the feature accrding the face/nn-face histgrams f the feature. A detectr can be ne r a cascade f such face/nnface strng classifiers, as in [19]. 4. Training f Individual Detectrs Hw t chse a gd cmbinatin f weak classifiers frm tens f thusands f features t cnstruct a pwerful detectr is a challenging prblem f feature selectin [8][15] and classifier design. We have devised a new bsting algrithm, called FlatBst [22], fr learning face detectrs fr the detectr-pyramid. Similar r better perfrmance than AdaBst is achieved with fewer weak classifiers. +1 dx' dx dy -2 y +1 x The detectrs in the pyramid are trained separately, using different training sets. An individual detectr is respnsible fr ne view, with pssible partial verlapping with its neighbring detectrs. Due t the symmetry f faces, we need t train side view detectrs fr ne-side nly, and mirrr the trained mdels fr the ther side. Fr ne feature used in left-side view, we mirrr its structure (See Figure 3) t cnstruct a new feature used fr right-side view. Each left-side view feature is mirrred by this way, and these new features are cmbined t cnstruct right side view detectrs. 5. Arbitrate amng Individual Outputs In ur framewrk, we have nine channels at the last layer; each channel represents ne facial view. T arbitrate amng these nine detectrs we use sme heuristic methds. Firstly, we cmbine the utput f sme view ranges int ne class. After cmbinatin, nine channels f view are cnverted t five channels (left prfile, left half-prfile, frntal, right half-prfile and right prfile). Fr example, we cmbine [ 90, 60 ] as left half-prfile. Then, we arbitrate utputs within these five view pses. We use Rwly's heuristic methd. We clean-up utputs f each detectr. See figure 4, A is the last utput f frnt face channel, and nly frntal faces are detected by this channel. B is the last utput f half-prfile channel. This channel in fact includes tw channels: right half-prfile channel and left half-prfile channel. Sme frntal faces will be detected by this channel because half-prfile detectrs will detect part f frntal face as half-prfile face (See Figure 5-B). C represents the last utput f prfile channel, and this channel includes tw channels: right prfile, left prfile t. A B C D Figure 4: Output f fntal (A), half-side (B), and full-side (C) view channels, and the final result (D) after pst-prcessing. Figure 3: Mirrring feature. On the left is a feature learned fr a left view detectr. On the right is the crrespnding feature mirrred fr the right view cunterpart. T arbitrate amng five channels, we present a nvel heuristic methd. In practice, we find half-prfile detectrs and prfile detectrs ften detect part f the frntal face as half-prfile r prfile face. S we prescribe that if a particular lcatin is identified as a frntal face, then all ther lcatins detected by prfile r half prfile face detectrs which verlap it are likely t be errrs, and can
4 therefre be eliminated. Similarly, if a particular lcatin is identified as half-prfile face, then all ther lcatins detected by prfile face detectrs are eliminated. D is the last utput arbitrating amng five channels (see figure 5). We can find that sme faces in B (detected by half-prfile channel) verlap part f faces in A (detected by frntal channel). We identify these faces in B which verlaps with faces in A as errrs, and eliminate them in the last utput (in D). By the same, half-prfile and prfile channels have sme verlaps t, and we eliminate faces detected by prfile channel, which verlap with faces by half-prfile channel. 6. Experimental Results This sectin describes the final face detectin system including training data preparatin, training prcedure, and the perfrmance cmparisn with previus view-based multi-view face detectin system. 6.1 Training Data Set Mre than 6,000 face samples are cllected by crpping frm varius surces (mstly frm vide). The view is in the range f > 90, 90 ] with 90 representing the left-side view and 0 representing the frntal view. A ttal number f abut 25,000 multi-view face images are generating frm the 6,000 samples by artificially shifting r rtatin. In ur system, we partitin multi-view face space int smaller and smaller (tp-dwn in the pyramid) subspaces f narrwer view ranges. At the tp layer, there is nly ne detectr. S all face sample are gruped int ne class. At the secnd layer, there are three detectrs, and face samples are gruped int three view classes (frntal, left-prfile and right-prfile). Face samples labeled with 20, 10, 0, 10, 20 are gruped as frntal faces, thse with > 90, 30 ] are gruped as left-prfile face and the faceswith > 30, 90 ] are gruped as right-prfile faces. At the third layer, there are nine detectrs, and face samples are gruped int nine view classes f [-90, -80], [-70, -60],, [80,90] degrees. 6.2 Training phase Figure 5: Multi-view face examples There are 13 detectrs in ur system, but we nly need train eight detectrs. The right view detectrs at the secnd and third levels can be cnstructed by mirrring features used in left view detectrs. This methd saves abut half training time fr ur system. These detectrs are trained separately, using their wn training data. Nn-face images used fr training these detectrs are cllected frm 12,000 images which dn t cntain face. Every detectr can be a cascade f strng classifiers and this guarantees high detectin speed. At the tp level, the detectr is trained using all the faces frm 90 t 90. It has a cascade f three strng classifiers structure. The number f features in these three strng classifiers is 5, 13 and 20 respectively. It can reject abut 50% nn-face training data, while retaining 99% face train data in training stage. At secnd level, there are three detectrs, each f which is trained t detect part range f the full-view faces. Training faces are separated int three classes t train these detectrs. At this level, each detectr has a cascade f six strng classifiers structure. In ur system, this level can ttally rejects abut 97% nn-face training data which g thrugh tp level, and retain 98% face train data in training stage. At bttm level, face training data is separated int nine classes. At this level, each detectr is a cascade f abut twenty strng classifiers structure. Each detectr has a detectin rate f abut 94%, and achieves a false psitive 6 rate f abut Detectin Results The final detectr is scanned acrss the image at multiple scales and lcatins. Scaling is achieved by scaling the detectrs themselves, rather than scaling the image. This prcess makes sense because the features can be evaluated at any scale with the same cst. We scale the detectrs using a factr f In Figure 4, the image is 320 by 240 pixel size. There are a ttal f 70,401 sub-windws t be verified in this image. The full-view detectr at the tp level needs 110 ms t prcess all these sub-windws. Abut 40% subwindws frm test image are rejected by this carse classifier, and nly 41,114 sub-windws can pass thrugh this classifier. At the secnd level, there are three detectrs. They ttally need 77 ms t prcess all the rest sub-windws. Abut 97% sub-windws f the 41,114 sub-windws are rejected by this level, and nly 1298 sub-windws pass thrugh this level. At the third level, there are nine detectrs. They prcess all these 1298 sub-windws. But they nly need 15 ms t d it, because mst sub-windws are rejected at first and secnd levels. The timing is summarized in Table 1. Level First Secnd Third Ttal Time 110ms 77ms 15ms 202ms Table 1: Times needed fr each level t run the 320*240 image. Because spend 15 ms is needed fr the third level, s it will nt affect the efficiency much f the whle system if we partitin multi-view face space int smaller subspaces f narrwer view ranges at the third level. That it t say (nw we have nine detectrs n the third level), if we decmpse multi-view face space int smaller subspaces (fr example:
5 19 view ranges), this system will still has high detectin speed, but the detectin rate will prbably be increased. Methd View-based Detectr-Pyramid Time 976ms 202ms Table 2: Cmparisn between the view-based and detectrpyramid architecture in speed fr multi-view face detectin. If we had nt adpted the pyramid-like framewrk presented in this paper, we can apply all these nine detectrs at the third level directly n all sub-windws withut carse classificatin at the tp and secnd levels. This methd will (we call it view-based) cst much time fr multi-view face detectin (see Table 2). Our system is tested n CMU prfile face test set. This test set cnsists f 208 images with 441 faces f which 347 were prfile views frm varius news web sites. These images were nt restricted in terms f subject matter r backgrund scenery. They were cllected frm varius news web sites. The database can be dwnladed at ml. We present sme results shwn in Fig 6. We als prvide a vide clip shwing multi-view face detectin at Figure 6: Examples f Detectin Results 7. Cnclusins In this paper, we have presented a detectr-pyramid architecture fr multi-view face detectin. Using a carse-tfine and simple-t-cmplex scheme, ur system slves the prblem effectively and efficiently by discarding mst f nn-face sub-windws using the simplest pssible features at the earliest pssible stage. This leads t the first real-time multi-view face detectin system. Given this framewrk demnstrates gd perfrmance in multi-view face detectin, we stress that the underlying architecture is fairly general and can be applied t ther appearance based bject detectin prblem. REFERENCE [1] A. Kuchinsky, C. Pering, M.L. Creech, D. Freeze, B. Serra and J. Gwizdka. Cnsumer Multimedia Organizatin and Retrieval System. Prceedings f ACM SIG CHI'99 Cnference. [2] A. P. Pentland, B. Mghaddam and T. Starner. Viewbased and mdular eigensapces fr face recgnitin. In CVPR, pages 84-91, [3] D. Rth, M.-H. Yang and N.Ahuja. A SNW-Based Face Detectr. NIPS'00. [4] E Alpaydin and C Kaynak. Cascading Classifiers. Kykernetika, 34(4), [5] E. Osuna, R. Freund, and F. Girsi. Training supprt vectr machines: An applicatin t face detectin. In CVPR, pages , [6] F. Crw. Summed-area tables fr texture mapping. In Prcessings f SIGGGRAPH, vlume 18(3), pages , [7] F. Fleuret and D. Geman. Carse-t-fine face detectin. Inter. Jurnal f Cmputer Visin, [8] G.H. Jhn, R. Khavi, and K. Pfleger. Irrelevant features and the subset selectin prblem. In Prcessings f the Eleventh Internatinal Cnference n Machine Learing, Pages ,1994. [9] H.A. Rwley, S.Baluja, and T.Kanade. Neural netwrk-based face detectin. IEEE Transactins n Pattern Analysis and Machine Intelligence, 20(1):23--28, [10] H. Schneiderman and T. Kanade. A statistical methd fr 3d bject detectin applied t faces and cars. In Prceedings f IEEE Cmputer Sciety Cnference n Cmputer Visin and Pattern Recgnitin, [11] J. Feraud, O. Bernier, and M. cllbert. A fast and accurate face detectr fr indexatin f face images. In Prc. Furth IEEE Cmputer Sciety Cnference n Cmputer Visin and Pattern Recgnitin, pages 52-59, [12] J. Friedman, T. Hastie, and R. Tibshirani. Lgistic regressin: a statistical view f bsting. Technical reprt, Department f Statistics, Sequia Hall, Stanfrd, Univerity, July [13] J. Huang, X. Sha, and H. Wechsler. Face pse discriminatin using supprt vectr machines (SVM). In
6 Prceedings f Internatinal Cnference Pattern Recgnitin, Brisbane, Queensland, Australia, [14] K.K. Sung and T. Pggi. Example-based learning fr view-based human face detectin. IEEE Transactins n Pattern Analysis and Machine Intelligence, 20(1):39--51, [15] K. Tieu and P. Vila. Bsting image retrival. In Prceedings f the IEEE Cnference n Cmputer Visin and Pattern Recgnitin, [16] L. Wisktt, J. Fellus, N. Kruger, and C. V. malsburg. Face recgnitin by elastic bunch graph matching. IEEE Transactins n Pattern Analysis and Machine Intelligence,19(7): , [17] M. Bichsel and A.P. Pentland. Human face recgnitin and the face image set s tplgy. In Image Understanding, Vlume 59, pages , [18] P. Pudil, J. Nvvicva, and J. Kittler. Flating search methds in feature selectin. Pattern Recgnitin Letters, , [19] P. Vila and M.J. Jnes, Rbust real-time bject detectin, IEEE ICCV Wrkshp n Statistical and Cmputatinal Theries f Visin. Vancuver, Canada. July 13, [20] R.E. Schapire and Y. Singer. Bsting algrithms using cnfidence-rated predictins. Prceedings f the Eleventh Annual Cnference n Cmputatinal Learning Thery, pages , [21] S. Gng, S.McKenna, and J.Cllins. An investigatin int face pse distributin. In Prc. IEEE Internatinal Cnference n Face and Gesture Recgnitin, Vermnt, [22] S.Z. Li, L. Zhu, Z.Q. Zhang, and H.J. Zhang. Statistical Learning f Multi-View Face Detectin. In Prc. 7 th Eurpean Cnference n Cmputer Visin. Cpenhagen, Demark. May 2002 [23] Y. Amit, D. Geman and K. Wilder. Jint inductin f shape features and tree classifiers. In IEEE Trans. Pattern Analysis. Mach. Intell, 19, , [24] Y. Freund and R.E. Schapire. A decisin-theretic generalizatin f nline learning and an applicatin t bsting. In cmputatin Learning Thery: Eurclt 95, pages 23-37, Springer-Verlag, 1995.
Using SPLAY Tree s for state-full packet classification
Curse Prject Using SPLAY Tree s fr state-full packet classificatin 1- What is a Splay Tree? These ntes discuss the splay tree, a frm f self-adjusting search tree in which the amrtized time fr an access,
More informationIntro to Machine Learning for Visual Computing
Intr t Machine Learning fr Visual Cmputing Drthea Tanning, Endgame Slides frm Derek Hiem, Peter Barnum CSC320: Intrductin t Visual Cmputing Michael Guerzhy Eamples f Categrizatin in Visin Part r bject
More informationUFuRT: A Work-Centered Framework and Process for Design and Evaluation of Information Systems
In: Prceedings f HCI Internatinal 2007 UFuRT: A Wrk-Centered Framewrk and Prcess fr Design and Evaluatin f Infrmatin Systems Jiajie Zhang 1, Keith A. Butler 2 1 University f Texas at Hustn, 7000 Fannin,
More informationTeaching Performance Evaluation Using Supervised Machine Learning Techniques
Teaching Perfrmance Evaluatin Using Supervised Machine Learning Techniques Elia Gergiana Dragmir University Petrleum-Gas f Pliesti, Department f Infrmatics Bd. Bucuresti Nr. 39, Pliesti, RO-100680, ROMANIA
More informationA FRAMEWORK FOR PROCESSING K-BEST SITE QUERY
Internatinal Jurnal f Database Management Systems ( IJDMS ) Vl., N., Octber 0 A FRAMEWORK FOR PROCESSING K-BEST SITE QUERY Yuan-K Huang * and Lien-Fa Lin Department f Infrmatin Cmmunicatin Ka-Yuan University;
More informationScatter Search And Bionomic Algorithms For The Aircraft Landing Problem
Scatter Search And Binmic Algrithms Fr The Aircraft Landing Prblem J. E. Beasley Mathematical Sciences Brunel University Uxbridge UB8 3PH United Kingdm http://peple.brunel.ac.uk/~mastjjb/jeb/jeb.html Abstract:
More informationComparing Boosted Cascades to Deep Learning Architectures for Fast and Robust Coconut Tree Detection in Aerial Images
Cmparing Bsted Cascades t Deep Learning Architectures fr Fast and Rbust Ccnut Tree Detectin in Aerial Images VISAPP2018, 27-29 January 2018 Steven Puttemans*, Kristf Van Beeck* and Tn Gedemé Intrductin
More informationHierarchical Classification of Amazon Products
Hierarchical Classificatin f Amazn Prducts Bin Wang Stanfrd University, bwang4@stanfrd.edu Shaming Feng Stanfrd University, superfsm@ stanfrd.edu Abstract - This prjects prpsed a hierarchical classificatin
More informationB Tech Project First Stage Report on
B Tech Prject First Stage Reprt n GPU Based Image Prcessing Submitted by Sumit Shekhar (05007028) Under the guidance f Prf Subhasis Chaudhari 1. Intrductin 1.1 Graphic Prcessr Units A graphic prcessr unit
More informationEscher s Circle Limit III
Escher s Circle Limit III Escher s Circle Limit III PCA: Principal Cmpnent Analysis The best pssible lwer dimensinal representatin based n linear prjectins. A basis f directins f variance rdered by their
More informationRetrieval Effectiveness Measures. Overview
Retrieval Effectiveness Measures Vasu Sathu 25th March 2001 Overview Evaluatin in IR Types f Evaluatin Retrieval Perfrmance Evaluatin Measures f Retrieval Effectiveness Single Valued Measures Alternative
More informationINSTALLING CCRQINVOICE
INSTALLING CCRQINVOICE Thank yu fr selecting CCRQInvice. This dcument prvides a quick review f hw t install CCRQInvice. Detailed instructins can be fund in the prgram manual. While this may seem like a
More informationEscher s Circle Limit III
Escher s Circle Limit III Escher s Circle Limit III ImageNet Images fr each categry f WrdNet 1000 classes 1.2mil images 100k test Tp 5 errr Dataset split Training Images Validatin Images Testing Images
More information$ARCSIGHT_HOME/current/user/agent/map. The files are named in sequential order such as:
Lcatin f the map.x.prperties files $ARCSIGHT_HOME/current/user/agent/map File naming cnventin The files are named in sequential rder such as: Sme examples: 1. map.1.prperties 2. map.2.prperties 3. map.3.prperties
More informationCS 309: Autonomous Intelligent Robotics. Instructor: Jivko Sinapov
CS 309: Autnmus Intelligent Rbtics Instructr: Jivk Sinapv http://www.cs.uteas.edu/~jsinapv/teaching/cs309_spring2017/ Machine Learning Annuncements Final Prject Presentatins Saturday, May 13, 7:00-10:00
More informationIWT TETRA TOBCAT Industrial applications of object categorization techniques. Steven Puttemans
Prgram 12:30 Registratin & cffee 13:00 Welcme and intrductin (Tn Gedemé) 13:15 Results frm the TOBCAT prject (Steven Puttemans) 14:00 Practical applicatins that use cmputer visin algrithms and their internal
More informationThese tasks can now be performed by a special program called FTP clients.
FTP Cmmander FAQ: Intrductin FTP (File Transfer Prtcl) was first used in Unix systems a lng time ag t cpy and mve shared files. With the develpment f the Internet, FTP became widely used t uplad and dwnlad
More informationInstance Based Learning
Instance Based Learning Vibhav Ggate The University f Texas at Dallas Readings: Mitchell, Chapter 8 surces: curse slides are based n material frm a variety f surces, including Tm Dietterich, Carls Guestrin,
More informationSoftware Engineering
Sftware Engineering Chapter #1 Intrductin Sftware systems are abstract and intangible. Sftware engineering is an engineering discipline that is cncerned with all aspects f sftware prductin. Sftware Prducts
More informationUsing CppSim to Generate Neural Network Modules in Simulink using the simulink_neural_net_gen command
Using CppSim t Generate Neural Netwrk Mdules in Simulink using the simulink_neural_net_gen cmmand Michael H. Perrtt http://www.cppsim.cm June 24, 2008 Cpyright 2008 by Michael H. Perrtt All rights reserved.
More informationCS 378 Computer Vision Problem set 4 Out: Thursday, Nov 5 Due: Tuesday, Nov 24, 11:59 PM. See the end of this document for submission instructions.
CS 378 Cmputer Visin Prblem set 4 Out: Thursday, Nv 5 Due: Tuesday, Nv 24, 11:59 PM See the end f this dcument fr submissin instructins. I. Shrt answer prblems [30 pints] 1. The SIFT descriptr is frmed
More informationChapter 6: Lgic Based Testing LOGIC BASED TESTING: This unit gives an indepth verview f lgic based testing and its implementatin. At the end f this unit, the student will be able t: Understand the cncept
More informationPerformance of VSA in VMware vsphere 5
Perfrmance f VSA in VMware vsphere 5 Perfrmance Study TECHNICAL WHITE PAPER Table f Cntents Intrductin... 3 Executive Summary... 3 Test Envirnment... 3 Key Factrs f VSA Perfrmance... 4 Cmmn Strage Perfrmance
More informationPages of the Template
Instructins fr Using the Oregn Grades K-3 Engineering Design Ntebk Template Draft, 12/8/2011 These instructins are fr the Oregn Grades K-3 Engineering Design Ntebk template that can be fund n the web at
More informationProcurement Contract Portal. User Guide
Prcurement Cntract Prtal User Guide Cntents Intrductin...2 Access the Prtal...2 Hme Page...2 End User My Cntracts...2 Buttns, Icns, and the Actin Bar...3 Create a New Cntract Request...5 Requester Infrmatin...5
More informationObject Detection with Statistical Template
Object Detectin with Statistical Template Cmputer Visin Jia-Bin Huang, Virginia Tech Many slides frm D. Hiem, J. Hays Administrative stuffs HW 5 is ut Due 11:59pm n Wed, Nvember 16 Scene categrizatin Please
More informationUiPath Automation. Walkthrough. Walkthrough Calculate Client Security Hash
UiPath Autmatin Walkthrugh Walkthrugh Calculate Client Security Hash Walkthrugh Calculate Client Security Hash Start with the REFramewrk template. We start ff with a simple implementatin t demnstrate the
More informationColored Image Watermarking Technique Based on HVS using HSV Color Model Piyush Kapoor 1, Krishna Kumar Sharma 1, S.S. Bedi 2, Ashwani Kumar 3, 1
ACEEE Int. J. n Netwrk Security, Vl. 02, N. 03, July 2011 Clred Image Watermarking Technique Based n HVS using HSV Clr Mdel Piyush Kapr 1, Krishna Kumar Sharma 1, S.S. Bedi 2, Ashwani Kumar 3, 1 Indian
More informationDate: October User guide. Integration through ONVIF driver. Partner Self-test. Prepared By: Devices & Integrations Team, Milestone Systems
Date: Octber 2018 User guide Integratin thrugh ONVIF driver. Prepared By: Devices & Integratins Team, Milestne Systems 2 Welcme t the User Guide fr Online Test Tl The aim f this dcument is t prvide guidance
More informationUsing the Swiftpage Connect List Manager
Quick Start Guide T: Using the Swiftpage Cnnect List Manager The Swiftpage Cnnect List Manager can be used t imprt yur cntacts, mdify cntact infrmatin, create grups ut f thse cntacts, filter yur cntacts
More informationOn the road again. The network layer. Data and control planes. Router forwarding tables. The network layer data plane. CS242 Computer Networks
On the rad again The netwrk layer data plane CS242 Cmputer Netwrks The netwrk layer The transprt layer is respnsible fr applicatin t applicatin transprt. The netwrk layer is respnsible fr hst t hst transprt.
More informationCntents 1 Intrductin Kit Cntents Requirements Installatin Gesture Sensr Kit Hardware and Jumper Settings De
Thin Film Pyrelectric IR Gesture Sensr Demnstratr Kit Fr lw pwer, high perfrmance gesture cntrl User Guide Versin 1.0 Dcument Revisin 1.00 20 th February 2012 Cntents 1 Intrductin... 3 1.1 Kit Cntents...
More informationSW-G using new DryadLINQ(Argentia)
SW-G using new DryadLINQ(Argentia) DRYADLINQ: Dryad is a high-perfrmance, general-purpse distributed cmputing engine that is designed t manage executin f large-scale applicatins n varius cluster technlgies,
More informationIRDS: Data Mining Process
IRDS: Data Mining Prcess Charles Suttn University f Edinburgh (many figures used frm Murphy. Machine Learning: A Prbabilistic Perspective.) Data Science Our wrking definitin Data science is the study f
More informationTwo-Dimensional Topology Structure between Vector Layers in GIS
2011, Scienceline Publicatin Jurnal f Civil Engineering and Urbanism Vlume 1, Issue 1: 10-14 (2011) ISSN-2252-0430 Tw-Dimensinal Structure between Vectr Layers in GIS Davd Parvinnezhad Hkmabadi* 1, Ali
More informationCMC Blade BIOS Profile Cloning
This white paper describes the detailed capabilities f the Chassis Management Cntrller s Blade BIOS Prfile Clning feature. Authr Crey Farrar This dcument is fr infrmatinal purpses nly and may cntain typgraphical
More informationObject Category Recognition
03/16/10 Object Categry Recgnitin Cmputer Visin CS 543 / ECE 549 University f Illinis Derek Hiem (Plus leftver material frm image categrizatin) Tday s class: categrizatin Mre abut classifiers Overview
More informationUsing the Swiftpage Connect List Manager
Quick Start Guide T: Using the Swiftpage Cnnect List Manager The Swiftpage Cnnect List Manager can be used t imprt yur cntacts, mdify cntact infrmatin, create grups ut f thse cntacts, filter yur cntacts
More informationMachine Learning Crash Course
Machine Learning Crash Curse Pht: CMU Machine Learning Department prtests G20 Cmputer Visin James Hays Slides: Isabelle Guyn, Erik Sudderth, Mark Jhnsn, Derek Hiem Dimensinality Reductin PCA, ICA, LLE,
More informationPriority-aware Coflow Placement and scheduling in Datacenters
Pririty-aware Cflw Placement and scheduling in Datacenters Speaker: Lin Wang Research Advisr: Biswanath Mukherjee Intrductin Cflw Represents a cllectin f independent flws that share a cmmn perfrmance gal.
More informationVisualizing High Dimensional Fuzzy Rules
21 Visualizing High Dimensinal Fuzzy Rules R. Hlve, M. R. Berthld, Berkeley/USA Abstract. In this paper we present an apprach t visualize a ptentially high-dimensinal and large number f (fuzzy) rules in
More informationEvolution of Regression I: From OLS to GPS to MARS
Evlutin f Regressin I: Frm OLS t GPS t MARS March 2013 Dan Steinberg Mikhail Glvnya Salfrd Systems Salfrd Systems 2013 1 Curse Outline Tday s Webinar Regressin Prblem quick verview Classical OLS the starting
More informationData Our project used three data sets that provide analysis on forty 15- to 20-minute course videos for CS144: Intro to Computer Networks :
Predicting Lecture Vide Cmplexity: Analysis f Supervised Regressin Nick Su Ismael Menjivar njsu@stanfrd.edu menjivar@stanfrd.edu December 8, 2014 Abstract In the past decade, use f Massively Open Online
More informationSEMANTIC IMAGE RETRIEVAL USING RELEVANCE FEEDBACK
SEMANTIC IMAGE RETRIEVAL USING RELEVANCE FEEDBACK Pushpa B. Patil 1 and Manesh Kkare 2 1 Dept. f Cmputer Engineering, BLDEA s, V.P.Dr. P. G. Halakatti CET. Bijapur, India Pushpa_ms@rediffmail.cm 2 Dept.
More informationUML : MODELS, VIEWS, AND DIAGRAMS
UML : MODELS, VIEWS, AND DIAGRAMS Purpse and Target Grup f a Mdel In real life we ften bserve that the results f cumbersme, tedius, and expensive mdeling simply disappear in a stack f paper n smene's desk.
More informationUiPath Automation. Walkthrough. Walkthrough Calculate Client Security Hash
UiPath Autmatin Walkthrugh Walkthrugh Calculate Client Security Hash Walkthrugh Calculate Client Security Hash Start with the REFramewrk template. We start ff with a simple implementatin t demnstrate the
More informationUPGRADING TO DISCOVERY 2005
Centennial Discvery 2005 Why Shuld I Upgrade? Discvery 2005 is the culminatin f ver 18 mnths wrth f research and develpment and represents a substantial leap frward in audit and decisin-supprt technlgy.
More informationStealing passwords via browser refresh
Stealing passwrds via brwser refresh Authr: Karmendra Khli [karmendra.khli@paladin.net] Date: August 07, 2004 Versin: 1.1 The brwser s back and refresh features can be used t steal passwrds frm insecurely
More informationQuick Guide on implementing SQL Manage for SAP Business One
Quick Guide n implementing SQL Manage fr SAP Business One The purpse f this dcument is t guide yu thrugh the quick prcess f implementing SQL Manage fr SAP B1 SQL Server databases. SQL Manage is a ttal
More informationReport Writing Guidelines Writing Support Services
Reprt Writing Guidelines Writing Supprt Services Overview The guidelines presented here shuld give yu an idea f general cnventins fr writing frmal reprts. Hwever, yu shuld always cnsider yur particular
More informationMacroFlo User Guide. IES Virtual Environment 6.5. Macroflo. VE 6.5 MacroFlo 1
MacrFl User Guide IES Virtual Envirnment 6.5 Macrfl VE 6.5 MacrFl 1 Cntents 1 Intrductin... 4 1.1 MacrFl View... 4 1.2 Overview f MacrFl View Interface Features... 4 1.2.1 Virtual Envirnment Menu Bar...
More informationAdaptive Learning Algorithm for SVM Applied to Feature Tracking
Adaptive Learning Algrithm fr SVM Applied t Feature Tracking Ashutsh Garg, Ira Chen, Thmas S. Huang University f Illinis at Urbana Champaign Email: {ashutsh,irachen,huang}@ifp.uiuc.edu Abstract The framewrk
More informationBecause this underlying hardware is dedicated to processing graphics commands, OpenGL drawing is typically very fast.
The Open Graphics Library (OpenGL) is used fr visualizing 2D and 3D data. It is a multipurpse pen-standard graphics library that supprts applicatins fr 2D and 3D digital cntent creatin, mechanical and
More informationHow to use DCI Contract Alerts
Hw t use DCI Cntract Alerts Welcme t the MyDCI Help Guide series Hw t use DCI Cntract Alerts In here, yu will find a lt f useful infrmatin abut hw t make the mst f yur DCI Alerts which will help yu t fully
More informationQuick Start Guide. Basic Concepts. DemoPad Designer - Quick Start Guide
Quick Start Guide This guide will explain the prcess f installing & using the DemPad Designer sftware fr PC, which allws yu t create a custmised Graphical User Interface (GUI) fr an iphne / ipad & embed
More informationPlease contact technical support if you have questions about the directory that your organization uses for user management.
Overview ACTIVE DATA CALENDAR LDAP/AD IMPLEMENTATION GUIDE Active Data Calendar allws fr the use f single authenticatin fr users lgging int the administrative area f the applicatin thrugh LDAP/AD. LDAP
More informationUser Manual for. Version: copyright by PHOENIX Showcontroller GmbH & Co.KG - Boris Bollinger GERMANY
User Manual fr Versin: 4.0 17.10.2011 cpyright by PHOENIX Shwcntrller GmbH & C.KG - Bris Bllinger GERMANY What s New Versin 2.0 This versin mainly aims t imprve the wrkflw and the utput f the raster image
More informationAn Approach to Recognize Bangla Digits from Digital Image
248 IJCSNS Internatinal Jurnal f Cmputer Science and Netwrk Security, VOL.11 N.6, June 2011 An Apprach t Recgnize Bangla Digits frm Digital Image Abdul Kadar Muhammad Masum 1, Mhammad Shahjalal 2, Md.
More informationCS510 Concurrent Systems Class 2. A Lock-Free Multiprocessor OS Kernel
CS510 Cncurrent Systems Class 2 A Lck-Free Multiprcessr OS Kernel The Synthesis kernel A research prject at Clumbia University Synthesis V.0 ( 68020 Uniprcessr (Mtrla N virtual memry 1991 - Synthesis V.1
More informationThe UNIVERSITY of NORTH CAROLINA at CHAPEL HILL
Yu will learn the fllwing in this lab: The UNIVERSITY f NORTH CAROLINA at CHAPEL HILL Cmp 541 Digital Lgic and Cmputer Design Spring 2016 Lab Prject (PART A): A Full Cmputer! Issued Fri 4/8/16; Suggested
More informationData Communications over Context-Based WMNs Delay Performance Evaluation
3rd IFIP/IEEE Internatinal Wrkshp n Bandwidth n Demand and Federatin Ecnmics (BD), C-lcated with NOMS Sympsium Osaka, Japan, 23rd April 2010 Data Cmmunicatins ver Cntext-Based WMNs Delay Perfrmance Evaluatin
More informationCHAPTER 8. Clustering Algorithm for Outlier Detection in. Data Mining
CHAPTER 8 Clustering Algrithm fr Outlier Detectin in Data Mining 8.1 Intrductin In many data mining applicatins, the primary step is detecting utliers in a dataset. Outlier detectin fr data mining is nrmally
More informationArius 3.0. Release Notes and Installation Instructions. Milliman, Inc Peachtree Road, NE Suite 1900 Atlanta, GA USA
Release Ntes and Installatin Instructins Milliman, Inc. 3424 Peachtree Rad, NE Suite 1900 Atlanta, GA 30326 USA Tel +1 800 404 2276 Fax +1 404 237 6984 actuarialsftware.cm 1. Release ntes Release 3.0 adds
More information1 Version Spaces. CS 478 Homework 1 SOLUTION
CS 478 Hmewrk SOLUTION This is a pssible slutin t the hmewrk, althugh there may be ther crrect respnses t sme f the questins. The questins are repeated in this fnt, while answers are in a mnspaced fnt.
More informationExtensible Query Processing in Starburst
Extensible Query Prcessing in Starburst Laura M. Haas, J.C. Freytag, G.M. Lhman, and H.Pirahesh IBM Almaden Research Center CS848 Instructr: David Tman Presented By Yunpeng James Liu Outline Intrductin
More informationThree-Dimensional Restricted-Orientation Convexity 1
Three-Dimensinal Restricted-Orientatin Cnvexity 1 Eugene Fink 2 Derick Wd 3 Abstract A restricted-rientatin cnvex set is a set f pints whse intersectin with lines frm sme fixed set is empty r cnnected.
More informationEntering an NSERC CCV: Step by Step
Entering an NSERC CCV: Step by Step - 2018 G t CCV Lgin Page Nte that usernames and passwrds frm ther NSERC sites wn t wrk n the CCV site. If this is yur first CCV, yu ll need t register: Click n Lgin,
More informationAppeared in the Proceedings of the 1997 DARPA Image Understanding Workshop, May, to accurately model the features as they appear
Appeared in the Prceedings f the 1997 DARPA Image Understanding Wrkshp, May, 1997 Parametric Feature Detectin Simn Baker y, Shree K. Nayar y, and Hirshi Murase z y Department f Cmputer Science, Clumbia
More informationDevelopment of a Robust Indoor 3D SLAM Algorithm
Develpment f a Rbust Indr 3D SLAM Algrithm Timthy Murphy Hnrs Tutrial Cllege Dr. David Chelberg Ohi University Schl f Electrical Engineering and Cmputer Science Russ Cllege f Engineering and Technlgy Prblem:
More informationThe QMF Family V Newsletter 3rd Quarter 2013 Edition
The QMF Family Newsletter 3rd Quarter 2013 Editin In This Issue Dive int QMF at the IBM Infrmatin On Demand Cnference irtual data surces and analytic queries in QMF A message frm the develpers f QMF: Changing
More informationMachine Learning Crash Course
Machine Learning Crash Curse Cmputer Visin Jia-Bin Huang, Virginia Tech Many slides frm D. Hiem, J. Hays Administrative stuffs HW 4 Due 11:59pm n Wed, Nvember 2 nd What is a categry? Why wuld we want t
More informationCustodial Integrator. Release Notes. Version 3.11 (TLM)
Custdial Integratr Release Ntes Versin 3.11 (TLM) 2018 Mrningstar. All Rights Reserved. Custdial Integratr Prduct Versin: V3.11.001 Dcument Versin: 020 Dcument Issue Date: December 14, 2018 Technical Supprt:
More informationWeb of Science Institutional authored and cited papers
Web f Science Institutinal authred and cited papers Prcedures written by Diane Carrll Washingtn State University Libraries December, 2007, updated Nvember 2009 Annual review f paper s authred and cited
More informationThe UNIVERSITY of NORTH CAROLINA at CHAPEL HILL
Yu will learn the fllwing in this lab: The UNIVERSITY f NORTH CAROLINA at CHAPEL HILL Designing a mdule with multiple memries Designing and using a bitmap fnt Designing a memry-mapped display Cmp 541 Digital
More informationRELEASE NOTES FOR PHOTOMESH 7.3.1
RELEASE NOTES FOR PHOTOMESH 7.3.1 Abut PhtMesh Skyline s PhtMesh fully autmates the generatin f high-reslutin, textured, 3D mesh mdels frm standard 2D phtgraphs, ffering a significant reductin in cst and
More informationInfrastructure Series
Infrastructure Series TechDc WebSphere Message Brker / IBM Integratin Bus Parallel Prcessing (Aggregatin) (Message Flw Develpment) February 2015 Authr(s): - IBM Message Brker - Develpment Parallel Prcessing
More informationGBT X-band (9 GHz): Pointing Stability
GBT X-band (9 GHz): Pinting Stability Dana S. Balser, Rnald J. Maddalena, Frank Ghig, & Glen I. Langstn 10 March 2002 Abstract The pinting stability f the GBT is evaluated by tracking the half-pwer pint
More informationHigh Security SaaS Concept Software as a Service (SaaS) for Life Science
Sftware as a Service (SaaS) fr Life Science Cpyright Cunesft GmbH Cntents Intrductin... 3 Data Security and Islatin in the Clud... 3 Strage System Security and Islatin... 3 Database Security and Islatin...
More informationLab 1 - Calculator. K&R All of Chapter 1, 7.4, and Appendix B1.2
UNIVERSITY OF CALIFORNIA, SANTA CRUZ BOARD OF STUDIES IN COMPUTER ENGINEERING CMPE13/L: INTRODUCTION TO PROGRAMMING IN C SPRING 2012 Lab 1 - Calculatr Intrductin In this lab yu will be writing yur first
More informationPAGE NAMING STRATEGIES
PAGE NAMING STRATEGIES Naming Yur Pages in SiteCatalyst May 14, 2007 Versin 1.1 CHAPTER 1 1 Page Naming The pagename variable is used t identify each page that will be tracked n the web site. If the pagename
More informationRelius Documents ASP Checklist Entry
Relius Dcuments ASP Checklist Entry Overview Checklist Entry is the main data entry interface fr the Relius Dcuments ASP system. The data that is cllected within this prgram is used primarily t build dcuments,
More informationof Prolog An Overview 1.1 An example program: defining family relations
An Overview f Prlg This chaptereviews basic mechanisms f Prlg thrugh an example prgram. Althugh the treatment is largely infrmal many imprtant cncepts are intrduced. 1.1 An example prgram: defining family
More informationCLOUD & DATACENTER MONITORING WITH SYSTEM CENTER OPERATIONS MANAGER. Course 10964B; Duration: 5 Days; Instructor-led
CENTER OF KNOWLEDGE, PATH TO SUCCESS Website: www.inf-trek.cm CLOUD & DATACENTER MONITORING WITH SYSTEM CENTER OPERATIONS MANAGER Curse 10964B; Duratin: 5 Days; Instructr-led WHAT YOU WILL LEARN This curse
More informationMultilevel Updating Method of Three- Dimensional Spatial Database Presented By: Tristram Taylor SE521
Multilevel Updating Methd f Three- Dimensinal Spatial Database Presented By: Tristram Taylr SE521 Written By: Yangting Liu, Gang Liu, Zhenwen He, Zhengping Weng Frm: China University f Gesciences Fr: 2010
More informationCreating a TES Encounter/Transaction Entry Batch
Creating a TES Encunter/Transactin Entry Batch Overview Intrductin This mdule fcuses n hw t create batches fr transactin entry in TES. Charges (transactins) are entered int the system in grups called batches.
More informationHP OpenView Performance Insight Report Pack for Quality Assurance
Data sheet HP OpenView Perfrmance Insight Reprt Pack fr Quality Assurance Meet service level cmmitments Meeting clients service level expectatins is a cmplex challenge fr IT rganizatins everywhere ging
More informationTutorial 5: Retention time scheduling
SRM Curse 2014 Tutrial 5 - Scheduling Tutrial 5: Retentin time scheduling The term scheduled SRM refers t measuring SRM transitins nt ver the whle chrmatgraphic gradient but nly fr a shrt time windw arund
More informationProject Extranet User Guide
Prject Extranet User Guide Drafted by: Francisc Galleg (fgalleg@ua.es) - UA Pedr Caselles (pcaselles@ua.es) - UA Raul Pamplega (rpamplega@ua.es) - UA 1 Table f cntents: 1. EXTRANET USER GUIDE 3 2. INTRODUCTION
More informationComputer Organization and Architecture
Campus de Gualtar 4710-057 Braga UNIVERSIDADE DO MINHO ESCOLA DE ENGENHARIA Departament de Infrmática Cmputer Organizatin and Architecture 5th Editin, 2000 by William Stallings Table f Cntents I. OVERVIEW.
More informationENERGY YIELD OF TRACKING PV SYSTEMS IN JORDAN
ENERGY YIELD OF TRACKING PV SYSTEMS IN JORDAN A. Al Tarabsheh, I. Etier, and A. Nimrat Hashemite University, Electrical Engineering Department, 13115 Zarqa, Jrdan. Abstract This paper analyzes the energy
More informationCisco Tetration Analytics, Release , Release Notes
Cisc Tetratin Analytics, Release 1.102.21, Release Ntes This dcument describes the features, caveats, and limitatins fr the Cisc Tetratin Analytics sftware. Additinal prduct Release ntes are smetimes updated
More informationCOP2800 Homework #3 Assignment Spring 2013
YOUR NAME: DATE: LAST FOUR DIGITS OF YOUR UF-ID: Please Print Clearly (Blck Letters) YOUR PARTNER S NAME: DATE: LAST FOUR DIGITS OF PARTNER S UF-ID: Please Print Clearly Date Assigned: 15 February 2013
More informationNetwork Rail ARMS - Asbestos Risk Management System. Training Guide for use of the Import Survey Template
Netwrk Rail ARMS - Asbests Risk Management System Training Guide fr use f the Imprt Survey Template The ARMS Imprt Survey Template New Asbests Management Surveys and their Survey Detail reprts can be added
More informationManaging Conflicts when Using Combination Strategies to Test Software
Managing Cnflicts when Using Cmbinatin Strategies t Test Sftware Mats Grindal Humanities and Infrmatics University f Skövde Sweden mats.grindal@his.se Jeff Offutt Inf. and Sftware Engineering Gerge Masn
More informationAutomated Canopy Estimator(ACE): Enhancing Crop Modelling and Decision Making in Agriculture. A. D. Coy, D.R. Rankine, M.A. Taylor, and D. C.
Autmated Canpy Estimatr(ACE): Enhancing Crp Mdelling and Decisin Making in Agriculture A. D. Cy, D.R. Rankine, M.A. Taylr, and D. C. Nielsen Outline Mtivatin Evaluatin Results Validatin Cnclusin Mtivatin
More informationCommon Language Runtime
Intrductin t.net framewrk.net is a general-purpse sftware develpment platfrm, similar t Java. Micrsft intrduced.net with purpse f bridging gap between different applicatins..net framewrk aims at cmbining
More informationSoil Image Segmentation and Texture Analysis: A Computer Vision Approach
Sil Image Segmentatin and Texture Analysis: A Cmputer Visin Apprach Bushra Nazir 1, Md. Iqbal Quraishi 2 U.G. Student, Department f Infrmatin Technlgy, Kalyani Gvernment Engineering Cllege, Kalyani, West
More informationAuthor guide to submission and publication
Authr guide t submissin and publicatin Cntents Cntents... 1 Preparing an article fr submissin... 1 Hw d I submit my article?... 1 The decisin prcess after submissin... 2 Reviewing... 2 First decisin...
More informationA solution for automating desktop applications with Java skill set
A slutin fr autmating desktp applicatins with Java skill set Veerla Shilpa (Senir Sftware Engineer- Testing) Mysre Narasimha Raju, Pratap (Test Autmatin Architect) Abstract LeanFT is a pwerful and lightweight
More informationSimple Regression in Minitab 1
Simple Regressin in Minitab 1 Belw is a sample data set that we will be using fr tday s exercise. It lists the heights & weights fr 10 men and 12 wmen. Male Female Height (in) 69 70 65 72 76 70 70 66 68
More information