The paper Mapping Auto-context Decision Forests to Deep ConvNets for Semantic Segmentation by David Richmond and colleagues received the prestigious Best Science Paper Award of the 27th British Machine Vision Conference (BMVC), held in York in September 2016. The paper was chosen out of 365 submissions. BMVC is among the top ranking international conferences in the area of computer vision.
The paper arose from a cooperation between Gene Myers' Lab at the MPI-CBG and the Computer Vision Lab Dresden of Prof. Carsten Rother at TU Dresden. The paper investigates the relation of the two most popular machine learning-based techniques for the task of semantic segmentation. The authors show that a mapping exists between both techniques which allows to combine their respective advantages, namely the versatility of neural networks and the efficient inference of random forests.