Bogdan Savchynskyy

A Comparative Study of Modern Inference Techniques for Structured Discrete Energy Minimization Problems

Szeliski et al. published an influential study in 2006 on energy minimization methods for Markov random fields. This study provided valuable insights in choosing the best optimization technique for certain classes of problems. While these insights remain generally useful today, the phenomenal success of random field models means that the kinds of inference problems that have to be solved changed significantly. Specifically, the models today often include higher order interactions, flexible connectivity structures, large label-spaces of different cardinalities, or learned energy tables. To reflect these changes, we provide a modernized and enlarged study. We present an empirical comparison of more than 27 state-of-the-art optimization techniques on a corpus of 2453 energy minimization instances from diverse applications in computer vision. To ensure reproducibility, we evaluate all methods in the OpenGM 2 framework and report extensive results regarding runtime and solution quality. Key insights from our study agree with the results of Szeliski et al. for the types of models they studied. However, on new and challenging types of models our findings disagree and suggest that polyhedral methods and integer programming solvers are competitive in terms of runtime and solution quality over a large range of model types.

A study of Nesterov's scheme for Lagrangian decomposition and MAP labeling

We study the MAP-labeling problem for graphical models by optimizing a dual problem obtained by Lagrangian decomposition. In this paper, we focus specifically on Nes-terovs optimal first-order optimization scheme for non-smooth convex programs, that has been studied for a range of other problems in computer vision and machine learning in recent years. We show that in order to obtain an efficiently convergent iteration, this approach should be augmented with a dynamic estimation of a corresponding Lip-schitz constant, leading to a runtime complexity of O(1/∊) in terms of the desired precision ∊. Additionally, we devise a stopping criterion based on a duality gap as a sound basis for competitive comparison and show how to compute it efficiently. We evaluate our results using the publicly available Middlebury database and a set of computer generated graphical models that highlight specific aspects, along with other state-of-the-art methods for MAP-inference.

A bundle approach to efficient MAP-inference by Lagrangian relaxation

Approximate inference by decomposition of discrete graphical models and Lagrangian relaxation has become a key technique in computer vision. The resulting dual objective function is convenient from the optimization point-of-view, in principle. Due to its inherent non-smoothness, however, it is not directly amenable to efficient convex optimization. Related work either weakens the relaxation by smoothing or applies variations of the inefficient projected subgradient methods. In either case, heuristic choices of tuning parameters influence the performance and significantly depend on the specific problem at hand. In this paper, we introduce a novel approach based on bundle methods from the field of combinatorial optimization. It is directly based on the non-smooth dual objective function, requires no tuning parameters and showed a markedly improved efficiency uniformly over a large variety of problem instances including benchmark experiments. Our code will be publicly available after publication of this paper.

InstanceCut: From Edges to Instances with MultiCut

This work addresses the task of instance-aware semantic segmentation. Our key motivation is to design a simple method with a new modelling-paradigm, which therefore has a different trade-off between advantages and disadvantages compared to known approaches. Our approach, we term InstanceCut, represents the problem by two output modalities: (i) an instance-agnostic semantic segmentation and (ii) all instance-boundaries. The former is computed from a standard convolutional neural network for semantic segmentation, and the latter is derived from a new instance-aware edge detection model. To reason globally about the optimal partitioning of an image into instances, we combine these two modalities into a novel MultiCut formulation. We evaluate our approach on the challenging CityScapes dataset. Despite the conceptual simplicity of our approach, we achieve the best result among all published methods, and perform particularly well for rare object classes.

Global Hypothesis Generation for 6D Object Pose Estimation

This paper addresses the task of estimating the 6D-pose of a known 3D object from a single RGB-D image. Most modern approaches solve this task in three steps: i) compute local features, ii) generate a pool of pose-hypotheses, iii) select and refine a pose from the pool. This work focuses on the second step. While all existing approaches generate the hypotheses pool via local reasoning, e.g. RANSAC or Hough-Voting, we are the first to show that global reasoning is beneficial at this stage. In particular, we formulate a novel fully-connected Conditional Random Field (CRF) that outputs a very small number of pose-hypotheses. Despite the potential functions of the CRF being non-Gaussian, we give a new, efficient two-step optimization procedure, with some guarantees for optimality. We utilize our global hypotheses generation procedure to produce results that exceed state-of-the-art for the challenging Occluded Object Dataset.

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.

Partial Optimality via Iterative Pruning for the Potts Model

We propose a novel method to obtain a part of an optimal non-relaxed integral solution for energy minimization problems with Potts interactions, known also as the minimal partition problem. The method empirically outperforms previous approaches likeMQPBO and Kovtun’s method in most of our test instances and especially in hard ones. As a starting point our approach uses the solution of a commonly accepted convex relaxation of the problem. This solution is then iteratively pruned until our criterion for partial optimality is satisfied. Due to its generality our method can employ any solver for the considered relaxed problem.

Probabilistic Correlation Clustering and Image Partitioning Using Perturbed Multicuts

We exploit recent progress on globally optimal MAP inference by integer programming and perturbation-based approximations of the log-partition function. This enables to locally represent uncertainty of image partitions by approximate marginal distributions in a mathematically substantiated way, and to rectify local data term cues so as to close contours and to obtain valid partitions. Our approach works for any graphically represented problem instance of correlation clustering, which is demonstrated by an additional social network example.

Evaluation of a first-order primal-dual algorithm for MRF energy minimization

We investigate the First-Order Primal-Dual (FPD) algorithm of Chambolle and Pock [1] in connection with MAP inference for general discrete graphical models. We provide a tight analytical upper bound of the stepsize parameter as a function of the underlying graphical structure (number of states, graph connectivity) and thus insight into the dependency of the convergence rate on the problem structure. Furthermore, we provide a method to compute efficiently primal and dual feasible solutions as part of the FPD iteration, which allows to obtain a sound termination criterion based on the primal-dual gap. An experimental comparison with Nesterovs first-order method in connection with dual decomposition shows superiority of the latter one in optimizing the dual problem. However due to the direct optimization of the primal bound, for small-sized (e.g. 20×20 grid graphs) problems with a large number of states, FPD iterations lead to faster improvement of the primal bound and a resulting faster overall convergence.

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.