A discriminatively trained, multiscale, deformable part model

Abstract

This paper describes a discriminatively trained, multiscale, deformable part model for object detection. Our system achieves a two-fold improvement in average precision over the best performance in the 2006 PASCAL person detection challenge. It also outperforms the best results in the 2007 challenge in ten out of twenty categories. The system relies heavily on deformable parts. While deformable part models have become quite popular, their value had not been demonstrated on difficult benchmarks such as the PASCAL challenge. Our system also relies heavily on new methods for discriminative training. We combine a margin-sensitive approach for data mining hard negative examples with a formalism we call latent SVM. A latent SVM, like a hidden CRF, leads to a non-convex training problem. However, a latent SVM is semi-convex and the training problem becomes convex once latent information is specified for the positive examples. We believe that our training methods will eventually make possible the effective use of more latent information such as hierarchical (grammar) models and models involving latent three dimensional pose.

Joint optimization of background subtraction and object detection for night surveillance

“…For the first-layer object detectors, we use histogram of oriented gradients (HOG) features [11] and apply the deformable-parts-based model in [7]. The deformable-parts-based model contains a mixture of components, allowing for better modeling of the variety of objects within a category.…”

“…The deformable-parts-model is trained discriminatively via a latent SVM. A detailed description of the model can be found in [7]. In our implementation, we first divide the training images into 2 groups based on the time period.…”

“…Table 1 and Table 2 give results for background subtraction and object detection respectively. Note that with one single model, our method outperforms a state-of-the-art background subtraction method in [6] and a state-of-the-art object detection method in [7]. Some visualized results from the proposed algorithm are given in Figure 3.…”

“…These observations suggest us to train some object detectors of the night scene in order to better find out the object region and ex- clude the lighted background. Since some parts of the object are more discriminative than others for recognizing the object, the part-based object model like [7] can be a reasonable choice for building an object detector.…”

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

“…there was no benefit to having unobserved hidden states) such as support vector machines (Oren et al 1997) and AdaBoost applied to face (Viola and Jones 2004) and text detection (Chen and Yuille 2005). More recent work-such as latent SVM (Felzenszwalb et al 2008) and boosting (Viola et al 2005)-does involve some hidden variables but these models still contain far less structure than the probabilistic grammars, such as AND/OR graphs, which seem necessary to deal with the full complexity of vision.…”