iccv2009 tutorial: boosting and random forest - part III

Björn Stenger 28 Sep 2009 2009 京都 Tutorial – Part 3 Tracking Using Classification and Online Learning

Roadmap Tracking by classification On-line Boosting Multiple Instance Learning Multi-Classifier Boosting Online Feature Selection Adaptive Trees Ensemble Tracking Online Random Forest Combining off-line & on-line Tracking by optimization

Tracking by Optimization Example: Mean shift tracking Given: target location in frame t , color distribution In frame t+1 : Minimize distance p candidate distribution q target distribution y location Mean shift: iterative optimization Finds local optimum Extension: downweight by background [Comaniciu et al. 00]

Support Vector Tracking [Avidan 01] Combines SVM classifier with optical flow-based tracking Input: Initial guess x of object location in frame t SVM classifier (trained on ~10,000 example images) Algorithm: Maximize SVM classification score SVM eqn Motion eqn (1 st order Taylor) Results in this task: Use first order Taylor approximation, obtain linear system Prior knowledge of classifier is used in tracking process, no online update!

Displacement Expert Tracking [Williams et al. 03] Learn a nonlinear mapping from images I to displacements δ u . Off-line training On-line tracking

Displacement Expert Tracking (2) [Williams et al. 03] Results on 136 frames of face sequence (no scale changes)

Online Selection of Discriminative Features [Collins et al. 03] Select features that best discriminate between object and background Feature pool: Discriminative score: measure separability (variance ratio) of fg/bg Within class variance should be small Total variance should be large

On-line Feature Selection (2) Input image: Feature ranking according to variance ratio [Collins et al. 03] Mean shift Mean shift Mean shift Median New location Combining estimates

Ensemble Tracking [Avidan 05] Use classifiers to distinguish object from background Image Feature space foreground background First location is provided manually All pixels are training data labeled {+1,-1} 11-dimensional feature vector 8 orientation histogram of 5x5 nhood 3 RGB values

Ensemble Tracking [Avidan 05] Confidence map Train T (=5) weak linear classifiers h : Combine into strong Classifier with AdaBoost Build confidence map from classifier margins Scale positive margin to [0,1] Mean shift Find the mode using mean shift Feature space foreground background

Ensemble Tracking Update [Avidan 05] Test examples x i using strong classifier H ( x ) For each new frame I j Run mean shift on confidence map Obtain new pixel labels y Keep K (=4) best (lowest error) weak classifiers Update their weights Train T-K (=1) new weak classifiers h 1 h 2 h 3 h 4 h 5

Ensemble Tracking Properties [Avidan 05] Adaptation to appearance changes Classification score as confidence measure Cannot handle full long-term occlusions (some by using particle filter) Adaptation vs drift trade-off Or “ Stability-plasticity dilemma” adaptive enough to allow learning new things while not diluting/forgetting previously learned patterns (too much) Could integrate off-line trained classifiers

AdaBoost (recap) [Freund, Schapire 97] Input Set of labeled training samples Weight distribution over samples for n=1 to N // number of weak classifiers train a weak classifier using samples and weight distribution calculate error calculate classifier weight update sample weights end Algorithm Result [slide credit H. Grabner] Feature space

AdaBoost (recap) [Freund, Schapire 97] h 2 h 3 h 4 h 1 H Training examples Weighted combination … ( x 1 , y 1 ) ( x N , y N ) 1/2 1/2

From Off-line to On-line Boosting [Oza, Russel 01] Input Input set of labeled training samples weight distribution over samples ONE labeled training sample strong classifier to update initial sample importance For n=1 to N train a weak classifier using samples and weight distribution calculate error calculate confidence update weight distribution End For n=1 to N update weak classifier using samples and importance update error estimate update confidence update importance End Algorithm Algorithm Off-line On-line [slide credit H. Grabner]

Online Boosting [Oza, Russell 01] Input ONE labeled training sample strong classifier for n=1 to N // number of weak classifiers update weak classifier using sample and importance update error estimate update classifier weight update sample importance end Algorithm Result Feature space initial sample importance [slide credit H. Grabner]

Online Boosting h 2 h 3 h 4 t h 1 ( x 1 , y 1 ) [Oza, Russell 01] ( x 2 , y 2 ) ( x 3 , y 3 ) ( x 4 , y 4 ) H Weighted combination Training example …

Convergence results [Oza 01] Convergence of classification function shown theoretically for naïve Bayes classifier experimentally: for lossless online classifiers (naïve Bayes, decision trees): agreement for lossy online (neural networks): sometimes substantial loss

Priming can help [Oza 01] Batch learning on first 200 points, then online

Online Boosting for Feature Selection [Grabner, Bischof 06] Each feature corresponds to a weak classifier Combination of simple features

Selectors [Grabner, Bischof 06] A selector chooses one feature/classifier from pool. Selectors can be seen as classifiers Classifier pool Idea: Perform boosting on selectors, not the features directly.

Online Feature Selection one sample Init importance Estimate errors Select best weak classifier Update weight Estimate importance Current strong classifier For each training sample [Grabner, Bischof 06] Global classifier pool Estimate errors Select best weak classifier Update weight Estimate errors Select best weak classifier Update weight Estimate importance

Tracking Principle [Grabner, Bischof 06] [slide credit H. Grabner]

Adaptive Tracking [Grabner, Bischof 06]

Limitations [Grabner, Bischof 06]

Multiple Instance Learning (MIL) Precisely labeled data is expensive Weakly labeled data is easier to collect Algorithm for allowing ambiguity in training data: Get bag of (data, label) pairs Bag is positive if one or more of its members is positive. [Keeler et al. 90, Dietterich et al. 97, Viola et al. 05]

Multiple Instance Learning Supervised learning training input MIL training input [Babenko et al. 09] Classifier MIL Classifier

Online MIL Boost [Babenko et al. 09] At time t get more training data 1 Update all candidate classifiers 2 Pick best K in a greedy fashion pool of weak classifier candidates

Online MIL Boost Frame t Frame t+ 1 Get data (bags) Update all classifiers in pool Greedily add best K to strong classifier [Babenko et al. 09]

Tracking Results [Babenko et al. 09]

On-line / Off-line Spectrum Tracking Detection General object/any background detector Fixed training set Object/Background classifier On-line update Adaptive detector Tracking with prior c/f Template Update Problem [Matthews et al. 04] Example strategies: Run detector in tandem to verify [Williams et al. 03] Include generative model [Woodley et al. 06][Grabner et al. 07] Integrate tracker and detector [Okuma et al. 04][Li et al. 07]

Semi-supervised Use labeled data as prior Estimate labels & sample importance for unlabeled data [Grabner et al. 08]

Tracking Results [Grabner et al. 08]

Beyond Semi-Supervised [Stalder et al. 09] Recognizer Object specific “ Adaptive prior” Updated by: pos: Tracked samples validated by detector neg: Background during detection “ too inflexible”

Learning to Track with Multiple Observers Observation Models Off-line training of observer combinations Optimal tracker for task at hand Labeled Training Data Idea: Learn optimal combination of observers (trackers) in an off-line training stage. Each tracker can be fixed or adaptive. Given: labeled training data, object detector [Stenger et al. 09]

Input: Set of observers Each returns a location estimate & confidence value [Stenger et al. 09] [OB] [LDA] [BLDA] [OFS] [MS] [C] [M] [CM] [BOF] [KLT] [FF] [RT] [NCC] [SAD] On-line boosting Linear Discriminant Analysis (LDA) Boosted LDA On-line feature selection On-line classifiers Color-based mean shift Color probability Motion probability Color and motion probability Histogram Block-based optical flow Kanade-Lucas-Tomasi Flocks of features Randomized templates Local features Normalized cross-correlation Sum of absolute differences Single template

Combination Schemes Find good combinations of observers automatically by evaluating all pairs/triplets (using 2 different schemes). 1) 2) [Stenger et al. 09]

How to Measure Performance? Run each tracker on all frames (don’t stop after first failure) Measure position error Loss of track when error above threshold Re-init with detector [Stenger et al. 09]

Results on Hand Data (Single Observers) [Stenger et al. 09]

Results on Hand Data Single observers Pairs of observers [Stenger et al. 09]

Tracking Results [Stenger et al. 09]

Face Tracking Results [Stenger et al. 09]

Multi-Classifier Boosting Simultaneously learn image clusters and classifiers [Kim et al. 09] AdaBoost Multi-class boosting with gating function

Online Multi-Class Boosting [Kim et al. 09] Handles multiple poses: take maximum classifier response

Online Adaptive Decision Trees [Basak 04] Sigmoidal soft partitioning function at each node hyperplane Activation value at node i Complete binary trees, tree structure is maintained, each class = subset of leaves, label leaf nodes beforehand For each training sample, adapt decision hyperplanes at all inner nodes via gradient descent on error measure (leaf node activation)

Adaptive Vocabulary Forests [Yeh et al. 07] Application: Efficient indexing, leafs represent visual words Batch learning: hierarchical k-means, cf. [Nister and Stewenius 06] [slide credit T. Yeh]

Incremental Building of Vocabulary Tree [Yeh et al. 07]

Tree Growing by Splitting Leaf Nodes [Yeh et al. 07]

Tree Adaptation with Re-Clustering [Yeh et al. 07] Identify affected neighborhood Remove exisiting boundaries Re-Cluster points

Accuracy drops when Adaptation is stopped [Yeh et al. 07] Recent accuracy T =100 R(j) = 1 if top ranked retrieved image belongs to same group

Tree Pruning [Yeh et al. 07] Limit the number of leaf nodes Keep record of inactivity period at each node, if limit reached, remove nodes with least-recently used access Allows for restructuring of heavily populated areas

On-line Random Forests [Saffari et al. 09] For each tree t Input: New training example Update tree t with k times Estimate Out-of-bag error end P(Discard tree t and insert new one) = Random forest …

Leaf Update and Split [Saffari et al. 09] Set of random split functions Split node when Number of samples in node > threshold 1 Gain of best split > threshold 2 class k Compute gain of each potential split function Current leaf node

Results [Saffari et al. 09] Convergence of on-line RF classification to batch solution on USPS data set Tracking error of online RF compared to online boosting

Conclusions On-line versions exist for Boosting and Random Forests Experimentally good convergence results (but few theoretical guarantees) Useful for Tracking via Classification A lot of code has been made available online by authors Detection – Tracking Spectrum Adaptation vs. Drift

References Avidan, S., Support Vector Tracking, IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), Hawaii, 2001. Avidan , S., Support Vector Tracking , IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), Vol. 26(8), pp. 1064--1072, 2004. Avidan, S., Ensemble Tracking, IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), Vol. 29(2), pp 261-271, 2007. Avidan, S., Ensemble Tracking, IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), San Diego, USA, 2005. Babenko, B., Yang, M.-H., Belongie, S., Visual Tracking with Online Multiple Instance Learning, Proc. CVPR 2009. Basak, J., Online adaptive decision trees, Neural Computation, v.16 n.9, p.1959-1981, September 2004. Collins, R. T., Liu, Y., Leordeanu, M., On-Line Selection of Discriminative Tracking Features, IEEE Transaction on Pattern Analysis and Machine Intelligence (PAMI), Vol 27(10), October 2005, pp.1631-1643. Collins, R. T., Liu, Y., On-Line Selection of Discriminative Tracking Features, Proceedings of the 2003 International Conference of Computer Vision (ICCV '03), October, 2003, pp. 346 - 352. Comaniciu, D., Ramesh, V., Meer, P., Kernel-Based Object Tracking, IEEE Trans. Pattern Analysis Machine Intell., Vol. 25, No. 5, 564-575, 2003. Comaniciu, D., Ramesh, V., Meer P., Real-Time Tracking of Non-Rigid Objects using Mean Shift, IEEE Conf. Computer Vision and Pattern Recognition, Hilton Head Island, South Carolina, Vol. 2, 142-149, 2000. T. G. Dietterich and R. H. Lathrop and T. Lozano-Perez, Solving the multiple instance problem with axis-parallel rectangles. Artificial Intelligence 89 31-71, 1997. Freund, Y. , Schapire, R. E. , A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1):119–139, August 1997. H. Grabner, C. Leistner, and H. Bischof, Semi-supervised On-line Boosting for Robust Tracking. In Proceedings European Conference on Computer Vision (ECCV), 2008. H. Grabner, P. M. Roth, H. Bischof, Eigenboosting: Combining Discriminative and Generative Information, IEEE Conference on Computer Vision and Pattern Recognition, 2007. H. Grabner, M. Grabner, and H. Bischof, Real-time Tracking via On-line Boosting, In Proceedings British Machine Vision Conference (BMVC), volume 1, pages 47-56, 2006. H. Grabner, and H. Bischof, On-line Boosting and Vision, In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), volume 1, pages 260-267, 2006. J. D. Keeler , D. E. Rumelhart , W.-K. Leow, Integrated segmentation and recognition of hand-printed numerals, Proc. 1990 NIPS 3, p.557-563, October 1990, Denver, Colorado, USA. T.-K. Kim and R. Cipolla, MCBoost: Multiple Classifier Boosting for Perceptual Co-clustering of Images and Visual Features, In Advances in Neural Information Processing Systems (NIPS), Vancouver, Canada, Dec. 2008. T-K. Kim, T. Woodley, B. Stenger, R. Cipolla, Online Multiple Classifier Boosting for Object Tracking, CUED/F-INFENG/TR631, Department of Engineering, University of Cambridge, June 2009.

Y. Li, H. Ai, S. Lao, M. Kawade, Tracking in Low Frame Rate Video: A Cascade Particle Filter with Discriminative Observers of Different Lifespans, Proc. CVPR, 2007. I. Matthews, T. Ishikawa, and S. Baker, The template update problem. In Proc. BMVC, 2003 I. Matthews, T. Ishikawa, and S. Baker, The Template Update Problem, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 26, No. 6, June, 2004, pp. 810 - 815. K. Okuma, A. Taleghani, N. De Freitas, J. Little, D. G. Lowe, A Boosted Particle Filter: Multitarget Detection and Tracking , European Conference on Computer Vision(ECCV), May 2004. Oza, N.C., Online Ensemble Learning, Ph.D. thesis, University of California, Berkeley. Oza, N.C. and Russell, S., Online Bagging and Boosting. In Eighth Int. Workshop on Artificial Intelligence and Statistics, pp. 105–112, Key West, FL, USA, January 2001. Oza, N.C. and Russell, S., Experimental Comparisons of Online and Batch Versions of Bagging and Boosting, The Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, California, 2001. Saffari, A., Leistner C., Santner J., Godec M., Bischof H., On-line Random Forests, 3rd IEEE ICCV Workshop on On-line Computer Vision, 2009. S. Stalder, H. Grabner, and L. Van Gool, Beyond Semi-Supervised Tracking: Tracking Should Be as Simple as Detection, but not Simpler than Recognition. In Proceedings ICCV’09 WS on On-line Learning for Computer Vision, 2009. B. Stenger, T. Woodley, R. Cipolla, Learning to Track With Multiple Observers. Proc. CVPR, Miami, June 2009. References & Code P. A. Viola and J. Platt and C. Zhang, Multiple instance boosting for object detection, Proceedings of NIPS 2005. O. Williams, A. Blake, and R. Cipolla, Sparse Bayesian Regression for Efficient Visual Tracking, in IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE Computer Society, August 2005. O. Williams, A. Blake, and R. Cipolla, A Sparse Probabilistic Learning Algorithm for Real-Time Tracking, in Proceedings of the Ninth IEEE International Conference on Computer Vision, October 2003. T. Woodley, B. Stenger, R. Cipolla, Tracking Using Online Feature Selection and a Local Generative Model, Proc. BMVC, Warwick, September 2007. T. Yeh, J. Lee, and T. Darrell, Adaptive Vocabulary Forests for Dynamic Indexing and Category Learning. Proc. ICCV 2007. Code: Severin Stalder, Helmut Grabner Online Boosting, Semi-supervised Online Boosting, Beyond Semi-Supervised Online Boosting http://www.vision.ee.ethz.ch/boostingTrackers/index.htm Boris Babenko MIL Track http://vision.ucsd.edu/~bbabenko/project_miltrack.shtml Amir Saffari http://www.ymer.org/amir/software/online-random-forests/

iccv2009 tutorial: boosting and random forest - part III

More Related Content

What's hot (20)

Similar to iccv2009 tutorial: boosting and random forest - part III (20)

More from zukun (20)

Recently uploaded (20)

iccv2009 tutorial: boosting and random forest - part III