Yet Another Computer Vision Index To Datasets (YACVID) - Details

Stand: 2017-08-20 000000m 11:57:58 - Overview

ok633
Attribute Current Content New
Name (Institute + Shorttitle)ECP Paris 2011 
Description (include details on usage, files and paper references)The ECP Paris 2011 dataset consists of 104 images taken from rue Monge in the fifth district of Paris, we kept only 20 for training and 10 for testing.

However, testing on 10 images is very limited and the Randomized Forest classifiers do not generalize very well on the whole training set, we process the data in the following way: for each image, we run 5 Q-learning agents, parsing the image for 5000 episodes with a data-driven exploration and the Randomized Forest based rewards. We compute the topological similarity and the detection rates of the 5 semantic segmentations found, and keep the one that maximizes the global detection rate (percentage of well labeled pixels). Then we show the confusion matrix on the 104 buildings, as well as the statistics of the topological similarities.

http://vision.mas.ecp.fr/Personnel/teboul/data.php 
URL Linkhttp://vision.mas.ecp.fr/Personnel/teboul/data.php 
Files (#)104 
References (SKIPPED)
Category (SKIPPED)Semantic Segmentation, Procedural Reconstruction 
Tags (single words, spaced)segmentation, semantic, procedural, reconstruction, urban, paris 
Last Changed2017-08-20 
Turing (2.12+3.25=?) :-)