Data Decomposition and Spatial Mixture Modeling for Part Based Model

Abstract

Abstract. This paper presents a system of data decomposition and spatial mixture modeling for part based models. Recently, many enhanced part based models (with e.g., multiple features, more components or parts) have been proposed. Nevertheless, those enhanced models bring high computation cost together with the risk of over-fitting. To tackle this problem, we propose a data decomposition method for part based models which not only accelerates training and testing process but also improves the performance on average. Besides, the original part based model uses a strict rigid structural model to describe the distribution of each part location. It is not "deformable" enough, especially for those instances with different viewpoints or poses in the same aspect ratio. To address this problem, we present a novel spatial mixture modeling method. The spatial mixture embedded model is then integrated into the proposed data decomposition framework. We evaluate our system on the challenging PASCAL VOC2007 and PASCAL VOC2010 datasets, demonstrating the state-of-the-art performance compared with other related methods in terms of accuracy and efficiency.

Recent Progress on Object Classification and Detection

“…To the best of our knowledge, the presented system achieves the state-of-the-art performance compared with all other related methods from both the competition and the open literature. Due to limit of space, we refer readers to [36,37] for details.…”

“…4. System framework used in VOC2010 [35] and VOC2011 [36] [ 36,37] in VOC2011 as shown in Fig. 4(b).…”

“…Object Representation: Representation models mainly include part based models [35,36,26,9,34] and rigid template models [8,28].…”

Combination features and models for human detection

“…"HSC" [25] denotes the Histogram of Sparse Codes. "DDSMM" [33] is a variation of part based models with data decomposition and spatial mixture modeling method. "CN-HOG" [19] is a HOG variant combined with color attributes.…”

Object Detection based on Deformable Parts Model with Global Context

“…Gentle Boost choose part of LBP feature and hog feature are combined and the object detection results have significantly improved. In addition, the model also introduces another important improvement by using a variety of contexts and RBF SVM to carry out contextual learning and make the average accuracy to reach 36.8%; besides, it introduced spatial hybrid modeling and contextual learningin 2011 [3][4].…”