Max Margin Learning of Hierarchical Configural Deformable Templates (HCDTs) for Efficient Object Parsing and Pose Estimation

Abstract

In this paper we formulate a hierarchical configurable deformable template (HCDT) to model articulated visual objects-such as horses and baseball players-for tasks such as parsing, segmentation, and pose estimation. HCDTs represent an object by an AND/OR graph where the OR nodes act as switches which enables the graph topology to vary adaptively. This hierarchical representation is compositional and the node variables represent positions and properties of subparts of the object. The graph and the node variables are required to obey the summarization principle which enables an efficient compositional inference algorithm to rapidly estimate the state of the HCDT. We specify the structure of the AND/OR graph of the HCDT by hand and learn the model parameters discriminatively by extending Max-Margin learning to AND/OR graphs. We illustrate the three main aspects of HCDTs-representation, inference, and learning-on the tasks of segmenting, parsing, and pose (configuration) estimation for horses and hu-

Representação de Gestos em Análise Sintática: uma Revisão Sistemática da Literatura

“…Os autores em [32] apresentam uma abordagem para reconhecimento de poses humanas e não-humanas (poses de cavalos). Nesse trabalho dezenas de pontos no contorno do corpo sob análise são identificados e organizados hierarquicamente, de forma que a cabeçaé o primeiro nível hierárquico e as extremidades são osúltimos, construindo o chamado Hierarchical Configurable Deformable Template.…”

“…A capacidade de determinar o contorno dos membros e objetos tem sido explorada também para utilizar a informacão do formato desses itens na construção de representações de dados. Em [32] a informação de contornoé utilizada para selecionar pontos ao redor do corpo ou do objeto sob análise e esses pontos são utilizados na representação dos dados a serem usados na análise de gestos.…”

Semantic part segmentation using compositional model combining shape and appearance

“…In terms of method, our work is related to [32,31], where they used compositional model for horse segmenta-tion. But they did not incorporate strong appearance cues into their compositional shape model, and they modeled only a few poses and viewpoints.…”

Joint Object and Part Segmentation Using Deep Learned Potentials

“…In the literature of segmentation, semantic parsing has also been actively investigated. However, due to its increased challenge in getting detailed boundaries, most previous work focused on parsing objects given both the semantic category and a cropped bounding box with no occlusion, such as human parsing [4,45,41,13,11], car parsing [35,13,28] or animal parsing [39]. Such methods are limited in their applications, as objects in realworld images are often occluded with large deformation and appearance variations, which is difficult to be handled by those shape-based [4] or appearance-based [11] models with hand-crafted features or bottom-up segments.…”