Representing Videos using Mid-level Discriminative Patches

1. Representing Videos using Mid-level Discriminative Patches CVPR2013 Poster

2. Outline • Introduction • Mining Discriminative Patches • Analyzing Videos • Experimental Evaluation & Conclusion

3. 1. Introduction • Q.1:What does it mean to understand this video ? • Q.2:How might we achieve such an understanding?

4. 1. Introduction • Video single feature vector semantic action object bits and pieces • General framework object detect primitive actions Bayesian networks storyline

5. 1. Introduction Drawback: computational models for identifying semantic entities are not robust enough to serve as a basis for video analysis

6. 1. Introduction use Discriminative spatio-temporal patches • Represent video not use global feature vector or set of semantic entities Discriminative spatio-temporal patches semantic object primitive human action correspond automatically mined human-object pair random but informative patches from training data consisting of hundreds of videos

7. 1. Introduction • spatio-temporal patches act as a discriminative vocabulary for action classification establish strong correspondence between patches in training and test videos. Using label transfer techniques align the videos and perform tasks (Ex. object localization, finer-level action detection etc.)

8. 1. Introduction

9. 1. Introduction

10. 2. Mining Discriminative Patches • Two conditions • Challenge (1)They occur frequently within a class. (2)They are distinct from patches in other classes. (1)Space of potential spatio-temporal patches is extremely large given that these patches can occur over a range of scales. (2)Overwhelming majority of video patches are uninteresting.

11. 2. Mining Discriminative Patches • Paradigm : bag-of words • Major drawbacks Step1:Sample a few thousand patches, perform k-means clustering to find representative clusters Step2:Rank these clusters based on membership in different action classes. (1)High-Dimensional Distance Metric (2)Partitioning

12. 2. Mining Discriminative Patches (1)High-Dimensional Distance Metric K-means use standard distance metric (Ex. Euclidean or normalized cross-correlation) Not well in high-dimensional spaces ※ We use HOG3D

13. 2. Mining Discriminative Patches (2)Partitioning Standard clustering algorithms partition the entire feature space. Every data point is assigned to one of the clusters during the clustering procedure. However, in many cases, assigning cluster memberships to rare background patches is hard. Due to the forced clustering they significantly diminish the purity of good clusters to which they are assigned

14. 2. Mining Discriminative Patches • Resolve these issues • Using Exemplar-SVM(e-SVM) to learn 1.Use an exemplar-based clustering approach 2.Every patch is considered as a possible cluster center Drawback : computationally infeasible Resolve: use motion  use Nearest Neighbor

15. 2. Mining Discriminative Patches Training partition: (form cluster) Training videos Validation partition : rank the clusters based on representativeness (ⅰ)Using simple nearest-neighbor approach(typically k=20) (ⅱ)Score each patch and rank (ⅲ)select a few patches per action class and use the e-SVM to learn (ⅳ)e-SVM are used to form clusters (ⅴ)re-rank

16. 2. Mining Discriminative Patches • Goal : smaller dictionary(set of representative patches) • Criteria (a)Appearance Consistency Consistency score (b)Purity tf-idf (score): same class/different class ※ All patches are ranked using a linear combination of the two score

17. 2. Mining Discriminative Patches

18. 3. Analyzing Videos • Action Classification • Beyond Classification: Explanation via Discriminative Patches input : test videos feature vector output : class Top n e-SVM detectors SVM classifier Q. How we can use detections of discriminative patches for establishing correspondences between training and test videos? Q. Which detections to select for establishing correspondence?

19. 3. Analyzing Videos • Context-dependent Patch Selection Vocabulary size : N candidate detections :{D1,D2,…,DN} whether or not the detection of e-SVM i is selected : xi Appearance term(Ai):e-SVM score for patch i Class Consistency term(Cli):This term promotes selection of certain e-SVMs over others given the action class. For example, for the weightlifting class it prefers selection of the patches with man and bar with vertical motion. We learn Clfrom the training data by counting the number of times that an e-SVM fires for each class.

20. 3. Analyzing Videos • Optimization Penalty term(Pij):is the penalty term for selecting apair of detections together. (1)e-SVMsiand j do not fire frequently together in the trainingdata. (2) the e-SVMs i and j are trained from differentaction classes. use Integer Program is an NP-hard problem 5~10 iterations IPFP algorithm

21. 4. Experimental Evaluation • Datasets :UCF-50,Olympics Sport • Implementation Details: • Classification Results ※Our current implementation considers only cuboid patches ※Patches are represented with HOG3D features (4x4x5 cells with 20 discrete orientations).

22. 4. Experimental Evaluation

23. 4. Experimental Evaluation

24. 4. Experimental Evaluation • Correspondence and Label Transfer

25. 4. Experimental Evaluation

26. 4. Experimental Evaluation • Conclusion 1.A new representation for videos . 2.Automatically mine these patches using exemplar-based clustering approach. 3.Obtaining strong correspondence and align the videos for transferring annotations. 4.As a vocabulary to achieve state of the art results for action classification.