Segmentation Results: VOC2010 BETA

Competition "comp5" (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.Entries equivalent to a selected submission are determined by bootstrapping the performance measure, and assessing if the differences between the selected submission and the others are not statistically significant (see sec 3.5 in VOC 2014 paper).

Average Precision (AP %)

  mean

aero
plane
bicycle

bird

boat

bottle

bus

car

cat

chair

cow

dining
table
dog

horse

motor
bike
person

potted
plant
sheep

sofa

train

tv/
monitor
submission
date
O2P_SVRSEGM_CPMC_CSI [?] 49.668.633.148.937.249.564.266.048.518.354.630.641.062.158.053.238.158.832.946.246.520-Nov-2012
CVC_Harmony [?] 35.456.720.631.033.920.857.651.435.87.128.122.624.329.349.437.823.337.618.145.630.729-Aug-2010
CVC_Harmony+Det [?] 40.158.323.139.037.836.463.262.431.99.136.824.629.437.560.644.930.136.819.444.135.930-Aug-2010
Svr-Segm [?] 39.752.527.432.334.547.460.654.842.69.032.925.327.132.447.138.336.850.321.935.240.930-Aug-2010
AHCRF [?] 30.331.018.819.523.931.353.545.324.48.231.016.415.827.348.131.131.027.519.834.826.430-Aug-2010
LSVM-MDPM [?] 31.836.723.920.918.841.062.749.021.58.321.17.016.428.242.540.519.633.613.334.148.526-Aug-2010
Bonn_FGT_Segm [?] 36.554.622.525.127.640.060.248.339.47.330.821.325.334.954.136.622.545.017.633.537.030-Aug-2010
REGION-LABEL [?] 29.138.821.513.69.231.151.944.425.76.726.012.512.831.041.944.45.737.510.033.232.330-Aug-2010
UC3M_Generative_Discriminative [?] 27.845.912.314.522.39.346.838.341.7-35.920.734.134.833.524.64.725.613.026.826.130-Aug-2010

Abbreviations

TitleMethodAffiliationContributorsDescriptionDate
Associative hierarchical CRFAHCRFOxford Brookes UniversityLubor Ladicky Christopher Russell Philip TorrAssociative hierarchical CRF with detectors and cooccurence. One pixel layer with dense feature boost potential, 6 segmentation with potentials based on histograms of features, detector potential based on part-based model sliding window detector, generatively trained cooccurence potential2010-08-30 22:14:09
FG Detection, FG TilingBonn_FGT_SegmUniveristy of BonnJoão Carreira, Adrian Ion, Fuxin Li, Cristian Sminchisescu- Detection on figure-ground segmentations; - Classification by detection; - Semantic segmentation on figure-ground tilings;2010-08-30 23:08:18
Harmony PotentialsCVC_HarmonyComputer Vision Center - Universitat Autònoma de BarcelonaJosep Maria Gonfaus, Xavier Boix, Fahad Kahn, Joost van de Weijer, Andrew Bagdanov, Marco Pedersoli, Jordi Gonzàlez, Joan Serrat Our submission is based on [1]. We use the CVC_flat classification submission as the observations for the global node. New Absolute Position Prior is added to the Superpixels Probabilities [1] Josep M. Gonfaus, Xavier Boix, Joost Van de Weijer, Andrew D. Bagdanov, Joan Serrat, and Jordi Gonzàlez, " Harmony Potentials for Joint Classification and Segmentation ", in Computer Vision and Pattern Recognition (CVPR), San Francisco, CA, 2010. 2010-08-29 20:10:26
CVC_Harmony plus Detection PriorsCVC_Harmony+DetComputer Vision Center - Universitat Autònoma de BarcelonaJosep Maria Gonfaus, Xavier Boix, Fahad Kahn, Joost van de Weijer, Andrew Bagdanov, Marco Pedersoli, Joan Serrat, Xavier Roca, Jordi GonzàlezOur submission is based on [1] and the previous submission "CVC_Harmony". Here we add the detection scores of [2] as location prior in the image. [1] J.M. Gonfaus, X. Boix, J. Van de Weijer, A. D. Bagdanov, J. Serrat, and J. Gonzàlez, "Harmony Potentials for Joint Classification and Segmentation", in CVPR 2010. [2] Felzenszwalb, Girshick, McAllester, Ramanan, "Object Detection with Discriminatively Trained Part Based Models", PAMI 20102010-08-30 23:52:37
LSVM Mixtures of deformable part modelsLSVM-MDPMUniversity of Chicago and TTI-CPedro Felzenszwalb (UChicago), Ross Girshick (UChicago), David McAllester (TTI-C)Our submission is based on an extension of [1,2]. Each model has 6 components with 8 parts. We associate a binary mask with each component to generate segmentations. The object detection models were trained from bounding boxes. The segmentation masks were trained from segmentations. [1] P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan. Object Detection with Discriminatively Trained Part Based Models. PAMI 32(9), Sept 2010 [2] http://people.cs.uchicago.edu/~pff/latent-release4/ 2010-08-26 04:07:34
O2P+SVRSEGM Regressor + Composite Statistical InferenceO2P_SVRSEGM_CPMC_CSI(1) Georgia Institute of Technology (2) University of California - Berkeley (3) Amazon Inc. (4) Lund University Fuxin Li (1), Joao Carreira (2), Guy Lebanon (3), Cristian Sminchisescu (4)We utilize a novel probabilistic inference procedure, Composite Statisitcal Inference (CSI) [1], on semantic segmentation using predictions on overlapping figure-ground hypotheses. Regressor predictions on segment overlaps to the ground truth object are modelled as generated by the true overlap with the ground truth segment plus noise, parametrized on the unknown percentage of each superpixel that belongs to the unknown ground truth. A joint optimization on all the superpixels and all the categories is then performed in order to maximize the likelihood of the SVR predictions. The optimization has a tight convex relaxation so solutions can be expected to be close to the global optimum. A fast and optimal search algorithm is then applied to retrieve each object. CSI takes the intuition from the SVRSEGM inference algorithm that multiple predictions on similar segments can be combined to better consolidate the segment mask. But it fully develops the idea by constructing a probabilistic framework and performing maximum composite likelihood jointly on all segments and categories. Therefore it is able to consolidate better object boundaries and handle hard cases when objects interact closely and heavily occlude each other. For each image, we use 150 overlapping figure-ground hypotheses generated by the CPMC algorithm (Carreira and Sminchisescu, PAMI 2012), SVRSEGM results, and linear SVR predictions on them with the novel second order O2P features (Carreira, Caseiro, Batista, Sminchisescu, ECCV2012; see VOC12 entry BONN_CMBR_O2P_CPMC_LIN) as the input to the inference algorithm. [1] Fuxin Li, Joao Carreira, Guy Lebanon, Cristian Sminchisescu. Composite Statistical Inference for Semantic Segmentation. CVPR 2013. 2012-11-20 18:39:06
Optimizing regions and their labelsREGION-LABELStanford UniversityM. Pawan Kumar, Stephen Gould, Haithem Turki, Dan Preston, Daphne KollerThe method groups pixels into regions and assigns the regions a unique semantic label simultaneously by minimizing a global energy function. Features used include those obtained from an object detector (deformable parts-based model) as well as a bag of SIFT words computed over the region. Inference as described in CVPR 2010, learning as described in ICCV 2009.2010-08-30 18:13:43
Svr-SegmSvr-SegmUniversity of BonnJoao Carreira, Fuxin Li, Adrian Ion, Cristian SminchisescuSupport vector regression to multiple descriptors extracted on segmentations. Descriptors include SIFT, color SIFT and HOG on foreground and background. Post-processing to eliminate spurious detections and segmentations. The winning method of 2009 challenge.2010-08-30 22:52:55
Combination of Generative Discriminative MethodsUC3M_Generative_DiscriminativeUniversidad Carlos III de MadridIván González-Díaz, Fernando Díaz de MaríaCombination of Supervised Topic Models with SVM-based discriminative methods for concurrent image recognition and segmentation2010-08-30 13:27:41