Dmitrij Schlesinger

Exact solution of permuted submodular minsum problems

In this work we show, that for each permuted submodular MinSum problem (Energy Minimization Task) the corresponding submodular MinSum problem can be found in polynomial time. It follows, that permuted submodular MinSum problems are exactly solvable by transforming them into corresponding submodular tasks followed by applying standart approaches (e.g. using MinCut-MaxFlow based techniques).

Unifying Registration and Segmentation for Multi-sensor Images

We propose a method for unifying registration and segmentation of multi-modal images assuming that the hidden scene model is a Gibbs probability distribution.

Inferring M-Best Diverse Labelings in a Single One

We consider the task of finding M-best diverse solutions in a graphical model. In a previous work by Batra et al. an algorithmic approach for finding such solutions was proposed, and its usefulness was shown in numerous applications. Contrary to previous work we propose a novel formulation of the problem in form of a single energy minimization problem in a specially constructed graphical model. We show that the method of Batra et al. can be considered as a greedy approximate algorithm for our model, whereas we introduce an efficient specialized optimization technique for it, based on alpha-expansion. We evaluate our method on two application scenarios, interactive and semantic image segmentation, with binary and multiple labels. In both cases we achieve considerably better error rates than state-of-the art diversity methods. Furthermore, we empirically discover that in the binary label case we were able to reach global optimality for all test instances.

Gibbs Probability Distributions for Stereo Reconstruction

A new approach for stereo reconstruction is proposed. This approach is based on a Gibbs probability distribution for surfaces in 3D space. The problem of stereo reconstruction is formulated then as a Bayes decision task. The main difference compared with known methods is the use of a more realistic cost function. In case of stereo reconstruction this function can be designed in some natural way, taking into account the properties of the surface model used. The proposed method solves the Bayes decision task approximately by a Gibbs Sampler. Learning of unknown distribution parameters is included as well, using the Expectation Maximization algorithm.

Combining Shape Priors and MRF-Segmentation

We propose a combination of shape prior models with Markov Random Fields. The model allows to integrate multiple shape priors and appearance models into MRF-models for segmentation. We discuss a recognition task and introduce a general learning scheme. Both tasks are solved in the scope of the model and verified experimentally.

A Probabilistic Segmentation Scheme

We propose a probabilistic segmentation scheme, which is widely applicable to some extend. Besides the segmentation itself our model incorporates object specific shading. Dependent upon application, the latter is interpreted either as a perturbation or as meaningful object characteristic. We discuss the recognition task for segmentation, learning tasks for parameter estimation as well as different formulations of shading estimation tasks.

Joint Training of Generic CNN-CRF Models with Stochastic Optimization

We propose a new CNN-CRF end-to-end learning framework, which is based on joint stochastic optimization with respect to both Convolutional Neural Network (CNN) and Conditional Random Field (CRF) parameters. While stochastic gradient descent is a standard technique for CNN training, it was not used for joint models so far. We show that our learning method is (i) general, i.e. it applies to arbitrary CNN and CRF architectures and potential functions; (ii) scalable, i.e. it has a low memory footprint and straightforwardly parallelizes on GPUs; (iii) easy in implementation. Additionally, the unified CNN-CRF optimization approach simplifies a potential hardware implementation. We empirically evaluate our method on the task of semantic labeling of body parts in depth images and show that it compares favorably to competing techniques.

Modelling composite shapes by Gibbs random fields

We analyse the potential of Gibbs Random Fields for shape prior modelling. We show that the expressive power of second order GRFs is already sufficient to express spatial relations between shape parts and simple shapes simultaneously. This allows to model and recognise complex shapes as spatial compositions of simpler parts.

Incremental learning in the energy minimisation framework for interactive segmentation

In this article we propose a method for parameter learning within the energy minimisation framework for segmentation. We do this in an incremental way where user input is required for resolving segmentation ambiguities. Whereas most other interactive learning approaches focus on learning appearance characteristics only, our approach is able to cope with learning prior terms; in particular the Potts terms in binary image segmentation. The artificial as well as real examples illustrate the applicability of the approach.

Joint Shape Classification and Labeling of 3-D Objects Using the Energy Minimization Framework

We propose a combination of multiple Conditional Random Field (CRF) models with a linear classifier. The model is used for the semantic labeling of 3-D surface meshes with large variability in shape. The model employs multiple CRFs of low complexity for surface labeling each of which models the distribution of labelings for a group of surfaces with a similar shape. Given a test surface the classifier exploits the MAP energies of the inferred CRF labelings to determine the shape class. We discuss the associated recognition and learning tasks and demonstrate the capability of the joint shape classification and labeling model on the object category of human outer ears.