Site Tools


Evidential Calibration

We provide the MATLAB® code for our evidential multiclass classifier calibration method.

  • calibTrain.m: train the calibration model given some validation data
  • score2prob.m: transform a vector of scores into probabilities
  • score2plaus.m: transform a vector of scores into plausibilities over singletons
Calibration code

Ph. Xu, F. Davoine, H. Zha and T. Denœux. Evidential calibration of binary SVM classifiers. International Journal of Approximate Reasoning (IJAR), Vol. 72, pages 55–70, May 2016.

Ph. Xu, F. Davoine and T. Denœux. Evidential Multinomial Logistic Regression for Multiclass Classifier Calibration. In Proceedings of the 18th International Conference on Information Fusion, pages 1106-1112, Washington, D.C., July 6-9, 2015.

Ph. Xu, F. Davoine and T. Denœux. Evidential Logistic Regression for Binary SVM Classifier Calibration. In F. Cuzzolin, editor, Belief Functions: Theory and Applications. Proceedings of the 3rd International Conference on Belief Functions, Springer, LNCS 8764, pages 49-57, Oxford, UK, September 26-28, 2014.
Paper Oral DOI BibTeX

Combination of pedestrian detectors

We provide the MATLAB® code for our evidential combination of pedestrian detectors.

To make the code work, you will need to download the Matlab evaluation/labeling code provided by the Caltech Pedestrian Detection Benchmark and Piotr's Matlab Toolbox. Copy the dbFusion.m file in the same directory as the dbEval.m file and simply run it. To divide the UsaTest dataset into a validation and testing set, add the following line to dbInfo.m:

case 'usaval' % Caltech Pedestrian Datasets (validation)
  setIds=6; subdir='USA'; skip=30; ext='jpg';
case 'usatest2' % Caltech Pedestrian Datasets (testing sub set)
  setIds=7:10; subdir='USA'; skip=30; ext='jpg';
  vidIds={0:11 0:10 0:11 0:11};
Combination code
Caltech evaluation code Link
Piotr's Matlab Toolbox Link

Ph. Xu, F. Davoine and T. Denoeux. Evidential Combination of Pedestrian Detectors. In Proceedings of the 25th British Machine Vision Conference (BMVC), Nottingham, UK, September 1-5, 2014. paper oral

KITTI semantic segmentation

A set of 107 images (70 for training and 37 for testing) from the KITTI Vision Benchmark Suite were manually annotated with the software Adobe® Photoshop® CS2. The left color images were annotated at the pixel level considering a set of 28 classes.

Training set Testing set
Ground truth
Left images
Right images
Velodyne data

For convenience, we also provide the left and right images, as well as the Velodyne data, associated to the ground truth annotations. These data were extracted from the raw sequences. They are copyright by the KITTI Vision Benchmark Suite and published under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License.


Ph. Xu, F. Davoine, J.-B. Bordes, H. Zhao and T. Denœux. Multimodal Information Fusion for Urban Scene Understanding. Machine Vision and Applications (MVA), Vol. 27, Issue 3, pages 331–349, April 2016.

Ph. Xu, F. Davoine, J.-B. Bordes, H. Zhao and T. Denoeux. Information Fusion on Oversegmented Images: An Application for Urban Scene Understanding. In Proceedings of the Thirteenth IAPR International Conference on Machine Vision Applications (MVA), pages 189-193, Kyoto, Japan, May 20-23, 2013. paperoral HAL

User Tools