立体匹配之(二):[MC-CNN] 2015CVPR: Stereo Matching by Training a Convolutional Neural Netw
1摘要project主页:https://github.com/jzbontar/mc-cnn基于patch的提取与比较,学习其相似性得到一个matchingcost,并将正确匹配的patch定义为正样本,其他为负样本。后处理包括:cross-basedcostaggregation,semiglobalmatching,aleft-rightconsistencycheck,subpixelen