A discriminatively trained, multiscale, deformable part model

TLDR

Deformable part models, though popular, had not yet shown value on challenging benchmarks like the PASCAL challenge, yet this system relies heavily on such parts. The authors aim to present a discriminatively trained, multiscale, deformable part model for object detection and anticipate that their training methods will enable richer latent models such as hierarchical grammars and 3D pose. They employ margin‑sensitive hard‑negative mining with a latent SVM framework, turning the non‑convex training into a semi‑convex problem once latent variables for positives are fixed. The model doubles average precision over the best 2006 PASCAL person detection results, outperforms the 2007 challenge in ten of twenty categories, and demonstrates that the latent SVM becomes convex when latent information is specified for positives.

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.