Better appearance models for pictorial structures

Abstract

We present a novel approach for estimating body part appearance models for pictorial structures. We learn latent relationships between the appearance of different body parts from annotated images, which then help in estimating better appearance models on novel images. The learned appearance models are general, in that they can be plugged into any pictorial structure engine. In a comprehensive evaluation we demonstrate the benefits brought by the new appearance models to an existing articulated human pose estimation algorithm, on hundreds of highly challenging images from the TV series Buffy the vampire slayer and the PASCAL VOC 2008 challenge.

Holistic Human Pose Estimation with Regression Forests

“…We evaluate our approach on it and compare with a part-based method [3]. For all the experiments, we use the PCP evaluation score [17].…”

“…The goal is to infer the most plausible body configuration given the image likelihoods, usually estimated by body part detectors, and a prior. One idea for improving the model is by using better appearance models [16][17][18]. This has also been done by using Random Forests for body part classification [19] or regression [20].…”

Has My Algorithm Succeeded? An Evaluator for Human Pose Estimators

“…We apply this evaluator learning framework to four recent publicly available methods: Eichner and Ferrari [5], Sapp et al [16], Andriluka et al [2] and Yang and Ramanan [22]. The algorithms are reviewed in section 2.…”

“…The auxiliary features, pose quality measure, and learning method are described in section 3. For the datasets, section 4, we use existing ground truth annotated datasets, such as ETHZ PASCAL Stickmen [5] and Humans in 3D [4], and supplement these with additional annotation where necessary, and also introduce a new dataset to provide a larger number of training and test examples. We assess the evaluator features and method on the four HPE algorithms, and demonstrate experimentally that the proposed evaluator can indeed predict when the algorithms succeed.…”

“…We describe the stages required to learn and test an evaluator, including the use of an annotated ground truth dataset for training and testing the evaluator (and we provide a new dataset for the HPE case), and the development of auxiliary features that have not been used by the (HPE) algorithm, but can be learnt by the evaluator to predict if the output is correct or not. Then an evaluator is built for each of four recently developed HPE algorithms using their publicly available implementations: Eichner and Ferrari [5], Sapp et al. [16], Andriluka et al [2] and Yang and Ramanan [22].…”