Action Classification Results: VOC2010 BETA

Competition "comp9" (train on VOC2010 data)

This leaderboard shows only those submissions that have been marked as public, and so the displayed rankings should not be considered as definitive.

Average Precision (AP %)

  mean

phoning

playing
instrument
reading

riding
bike
riding
horse
running

taking
photo
using
computer
walking

submission
date
OppponentHOP [?] 73.259.654.242.084.990.483.237.170.676.123-Jun-2013
Multikernel+KDA [?] 68.352.653.535.981.089.386.532.859.268.730-Aug-2010
SVM-DOSP-MULTFEATS [?] 67.247.057.826.978.889.787.332.560.070.130-Aug-2010
CVC-BASE [?] 66.956.256.534.775.183.686.525.460.069.229-Aug-2010
CVC-SEL [?] 66.449.852.834.374.285.585.125.064.172.530-Aug-2010
SPM+HT [?] 66.153.253.630.278.288.484.630.460.961.929-Aug-2010
SVM_WHGO_SIFT_CENTRIST_low-level modeling [?] 62.347.247.924.574.281.079.524.958.671.523-Aug-2010
dhog-ksvm [?] 59.753.543.032.067.968.883.034.145.960.430-Aug-2010
a * SVM-SIFT + (1-a) * LSVM [?] 58.249.237.722.273.277.181.724.353.756.929-Aug-2010
SVM-SIFT [?] 52.947.929.121.753.576.778.326.042.956.429-Aug-2010
LSVM [?] 48.540.429.932.253.562.273.617.645.841.529-Aug-2010

Abbreviations

TitleMethodAffiliationContributorsDescriptionDate
SVM classifier with multiple featuresCVC-BASEComputer Vision Center, Universitat Autonoma de Barcelona, SpainNataliya Shapovalova, Wenjuan Gong, Fahad Shahbaz Khan, Josep M. Gonfaus, Marco Pedersoli, Andrew D. Bagdanov, Joost van de Weijer, Jordi GonzálezBaseline CVC submission for action recognition. Standard BoW model over multiple features including PHOG, grayscale SIFT and (various) color SIFT descriptors. Foreground/background modeled separately, spatial pyramid over several features for foreground representation. Late fusion of feature-specific SVM outputs for final action score.2010-08-29 18:01:14
SVM classifier with per-class feature selectionCVC-SELComputer Vision Center, Universitat Autonoma de Barcelona, SpainNataliya Shapovalova, Wenjuan Gong, Fahad Shahbaz Khan, Josep M. Gonfaus, Marco Pedersoli, Andrew D. Bagdanov, Joost van de Weijer, Jordi GonzálezEnhanced CVC submission built upon CVC-BASE for action recognition. Standard BoW model over multiple features from CVC-BASE plus contextual object descriptors. Cross-validation procedure for action-specific feature and kernel selection. Foreground/background/neighborhood modeled separately, spatial pyramid over several features for foreground representation. Object detection based on deformable part-based detector incorporated. Late fusion of feature-specific SVM outputs for final action score.2010-08-30 21:16:14
Felzenszwalb's part-based model.LSVMFrance, INRIA - Willow ProjectVincent Delaitre, Ivan Laptev, Josef SivicFelzenszwalb's part-based model trained on full train+val set with default parameters.2010-08-29 13:57:58
Mulitkernel fusion with KDAMultikernel+KDA The University of SurreyPiotr Koniusz, Muhammad Atif Tahir, Mark Barnard, Fei Yan, Krystian MikolajczykKernel-level fusion with Spatial Pyramid Grids, Soft Assignment and Kernel Discriminant Analysis using spectral regression. 18 kernels have been generated from 18 variants of SIFT.2010-08-30 01:48:44
Opponent late fusion.OppponentHOPUniversity of SurreyPiotr KoniuszThe opponent SIFT is extracted on the dense grid and combined with Spatial Coordinate Coding, then Sparse Coding is applied. Advanced Pooling and Statistics are extracted from the features. Kernel Discriminant Classifier is applied on the linear kernel.2013-06-23 19:52:21
Spatial Pyramids and Hough TransformSPM+HTINRIANorberto Adrián Goussies, Arnau Ramisa, Cordelia SchmidSpatial Pyramids on the bounding box, on the image and a hough transform for taking into account the object-person interactions for bicycle, horse and tvmonitor. Trained on trainval with 5-fold cross-validation.2010-08-29 02:53:04
SVM & dense saptial pyramid w/ multiple featuresSVM-DOSP-MULTFEATSUniversity of Caen GREYC and INRIA LEARGaurav Sharma, Frederic Jurie, Cordelia SchmidMultiple chi squared kernels are computed: spatial pyramid (SP) w/ dense SIFT, dense overlapping SP w/ HOG, texture filter, LAB values (bag-of-words w/ the above features) and edge dir hists. They are computed on full images, person bounding boxes (BB) and BB of the lower part (simple stretch-scale of person BB) expected to contain horse, bike etc. They are combined with class specific binary weights based on their perf on val set. Finally, class specific SVMs trained on train+val.2010-08-30 16:21:39
SVM classifier with dense SIFT features.SVM-SIFTFrance, INRIA - Willow ProjectVincent Delaitre, Ivan Laptev, Josef SivicSVM-Light classifier with dense SIFT features. Trained using 5-fold cross-validation. Re-trained on full train+val set with fixed parameters2010-08-29 13:47:55
SVM classifier on low-level modeling based approacSVM_WHGO_SIFT_CENTRIST_low-level modelingDepartment of Automatic Control, College of Mechatronics and Automation, National University of Defense TechnologyLi Zhou, Zongtan Zhou, Dewen HuThis method is based on a low-level modeling strategy. The approach works by creating multiple resolution images and partitioning them into sub-regions at different scales. We represent each sub-region with WHGO, CENTRIST and SIFT descriptors and combine the features of different descriptor and resolution channels through an SVM classifier to form the final decision function. 2010-08-23 06:02:40
Combination of SVM and DPM with learned weights.a * SVM-SIFT + (1-a) * LSVMFrance, INRIA - Willow ProjectVincent Delaitre, Ivan Laptev, Josef SivicCombination of a SVM classifier(with dense SIFT features, trained using 5-fold cross-validation and re-trained on full train+val set with fixed parameters) and of the Felzenszwalb's deformable part-based model (trained on full train+val set with default parameters).The classification score is obtained by a linear combination of the scores of the two classifiers:the two classifiers are also trained on the train set and the weights of this combination are determined by optimizing over the val set.2010-08-29 14:15:02
kernel svm classifier with dhog featuredhog-ksvmUniversity of Missouri - ColumbiaXutao Lv, Xiaoyu Wang, Xi Zhou, Tony X. Hantrain SVM model with different kernels on dhog feature.2010-08-30 22:08:53