Improving Detection of Deformable Objects in Volumetric Data
British Machine Vision Conference (BMVC), 2014
Abstract: In this paper, we investigate class level object detection of deformable objects. To this end, we aim for cell detection in volumetric images of dense plant tissue (Arabidopsis Thaliana), obtained from a confocal laser scanning micro scope. In 3D volumetric data, the detection model does not have to deal with scale, occlusion and viewpoint dependent changes of the appearance, however, our application needs high recall and precision. We implement Felsenszwalb’s Deformable Part Model for volumetric data. Corresponding locations for part training are obtained via elastic registration. We identify limitations of its star shaped deformation model and show that a pairwise connected detection model can outperform the star shaped Deformable Part Model in this setting.
Publisher's link
Images and movies
BibTex reference
@InProceedings{MR14, author = "D. Mai and J. D{\"u}rr and K. Palme and O. Ronneberger", title = "Improving Detection of Deformable Objects in Volumetric Data", booktitle = "British Machine Vision Conference (BMVC)", year = "2014", note = "http://www.bmva.org/bmvc/2014/papers/paper056/index.html", url = "http://lmb.informatik.uni-freiburg.de/Publications/2014/MR14" }