DeepCut is a method for multi-person pose estimation that jointly solves detection and pose estimation tasks. It infers the number of people in a scene, identifies occluded body parts, and disambiguates body parts between people in close proximity. Unlike previous two-stage approaches that first detect people and then estimate their poses, DeepCut uses a joint formulation based on partitioning and labeling body-part hypotheses generated by CNN-based part detectors. This formulation is an instance of an integer linear program (ILP) that implicitly performs non-maximum suppression on part candidates and groups them into configurations respecting geometric and appearance constraints. Experiments on four datasets show state-of-the-art results for both single and multi-person pose estimation. The method addresses key challenges in multi-person pose estimation, including partial visibility, significant overlap of bounding boxes, and the unknown number of people in an image. It formulates the problem as an ILP, allowing for an unknown number of people and effectively performing non-maximum suppression. The formulation is cast as an ILP, enabling the computation of bounds and feasible solutions with a certified optimality gap. The method uses two CNN variants to generate body part candidates, which, combined with the model, achieve state-of-the-art results. The paper also discusses related work, including previous approaches to pose estimation and multi-person pose estimation. It highlights the advantages of the proposed method, including its ability to handle occlusions, perform non-maximum suppression, and use an ILP formulation. The method is evaluated on several datasets, including LSP, MPII, WAF, and MPII Multi-Person, showing significant improvements over previous methods. The results demonstrate that the joint formulation is crucial for disambiguating multiple and potentially overlapping persons. The method is available at http://pose.mpi-inf.mpg.de.DeepCut is a method for multi-person pose estimation that jointly solves detection and pose estimation tasks. It infers the number of people in a scene, identifies occluded body parts, and disambiguates body parts between people in close proximity. Unlike previous two-stage approaches that first detect people and then estimate their poses, DeepCut uses a joint formulation based on partitioning and labeling body-part hypotheses generated by CNN-based part detectors. This formulation is an instance of an integer linear program (ILP) that implicitly performs non-maximum suppression on part candidates and groups them into configurations respecting geometric and appearance constraints. Experiments on four datasets show state-of-the-art results for both single and multi-person pose estimation. The method addresses key challenges in multi-person pose estimation, including partial visibility, significant overlap of bounding boxes, and the unknown number of people in an image. It formulates the problem as an ILP, allowing for an unknown number of people and effectively performing non-maximum suppression. The formulation is cast as an ILP, enabling the computation of bounds and feasible solutions with a certified optimality gap. The method uses two CNN variants to generate body part candidates, which, combined with the model, achieve state-of-the-art results. The paper also discusses related work, including previous approaches to pose estimation and multi-person pose estimation. It highlights the advantages of the proposed method, including its ability to handle occlusions, perform non-maximum suppression, and use an ILP formulation. The method is evaluated on several datasets, including LSP, MPII, WAF, and MPII Multi-Person, showing significant improvements over previous methods. The results demonstrate that the joint formulation is crucial for disambiguating multiple and potentially overlapping persons. The method is available at http://pose.mpi-inf.mpg.de.