Beyond Physical Connections: Tree Models in Human Pose Estimation

By Fang Wang, Yi Li. 2013

155 citations recorded by the source corpus.

Simple tree models for articulated objects prevails in the last decade. However, it is also believed that these simple tree models are not capable of capturing large variations in many scenarios, such as human pose estimation. This paper attempts to address three questions: 1) are simple tree models sufficient? more specifically, 2) how to use tree models effectively in human pose estimation? and 3) how shall we use combined parts together with single parts efficiently? Assuming we have a set of single parts and combined parts, and the goal is to estimate a joint distribution of their locations. We surprisingly find that no latent variables are introduced in the Leeds Sport Dataset (LSP) during learning latent trees for deformable model, which aims at approximating the joint distributions of body part locations using minimal tree structure. This suggests one can straightforwardly use a mixed representation of single and combined parts to approximate their joint distribution in a simple tree model. As such, one only needs to build Visual Categories of the combined parts, and then perform inference on the learned latent tree. Our method outperformed the state of the art on the LSP, both in the scenarios when the training images are from the same dataset and from the PARSE dataset. Experiments on animal images from the VOC challenge further support our findings.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Recent advances in convolutional neural networks (2015, background): “Their classiﬁcation pipeline is composed of three types of attentions: the bottom-up attention proposes candidate patches, the object-level top-down attention selects relevant patches of a certain object, and the part-level top-down attention localizes discriminative parts.”
DeeperCut: A Deeper, Stronger, and Faster Multi-person Pose Estimation Model (2016, background): “Various formulations have been proposed based on tree [13,14,15,6] and non-tree models [16,17].”
DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation (2015, methodology): “Finally, all proposed approaches signiﬁcantly outperform earlier non-deep learning based methods [40, 28] relying on hand-crafted image features.”
Recurrent Human Pose Estimation (2016, neutral): “ LSP, 1 it.) 95.6 88.8 80.7 75.5 83.0 86.2 80.6 84.3 Ours (MPII & LSP, 2 it.) 95.2 89.0 81.5 77.0 83.7 87.0 82.8 85.2 Pishchulin et al. [28] 87.2 56.7 46.7 38.0 61.0 57.5 52.7 57.1 Wang Li et al. [42] 84.7 57.1 43.7 36.7 56.7 52.4 50.8 54.6 Carreira et al. [9] 90.5 81.8 65.8 59.8 81.6 70.6 62.0 ”
Body Parts Dependent Joint Regressors for Human Pose Estimation in Still Images (2014, background): “For instance, a body part can be assigned with high confidence to two nodes of the pictorial structure model in case of weak part templates or occlusions, e.g., as illustrated in Figure 1, the left and right body part are sometimes assigned to a single observation.”
TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 1 ModDrop (2018, unknown): “Finally, graphical models, exploring spatial relationships between body and hand parts, have recently attracted close attention [28], [29].”
Latent Regression Forest: Structured Estimation of 3D Articulated Hand Posture (2014, methodology): “Whereas Wang and Li [20] use a Latent Tree Model to approximate the joint distributions of body part locations.”
Multi-source Deep Learning for Human Pose Estimation (2014, methodology): “Despite many years of research [49, 51, 1, 39, 5, 54, 53], pose estimation remains a difficult problem.”
Multi-task Recurrent Neural Network for Immediacy Prediction (2015, methodology): “Existing works either considered only single person pose estimation [31, 34, 8, 39, 6, 9, 30, 24, 2, 26, 14, 29, 35, 25, 1, 21] or multi-person pose estimation [37, 13], while our model jointly takes 1-person and 2-person pose estimation as the input.”
Structured Feature Learning for Pose Estimation (2016, methodology): “The first is to model the geometrical distribution of body joints [35, 30, 31, 19, 9, 6, 1, 11, 25, 2, 21, 17, 34, 7] which can be viewed as post processing on detection score maps and prediction labels.”
Combining local appearance and holistic view: Dual-Source Deep Neural Networks for human pose estimation (2015, methodology): “Based on the appearance models built for each part, these methods usually leverage tree-structured graphical models to further impose the pairwise geometric constraints between parts [2, 3, 8, 7, 21].”
Multi-scale Deep Learning for Gesture Detection and Localization (2014, background): “els, exploring spatial relationships between body and hand parts, have recently attracted close attention from the vision community [25, 26].”
Knowledge-Guided Deep Fractal Neural Networks for Human Pose Estimation (2017, unknown): “CVPR’13 [74], 84.”
Mirror, mirror on the wall, tell me, is the error small? (2015, methodology): “The one by Wang and Li [21] introduced latent variables in tree model learning that led to improvements.”
Multi-person Pose Estimation with Local Joint-to-Person Associations (2016, methodology): “ 61.8 76.9 10 Table 1: Pose estimation results (AP) on the validation test set (1200 images) of the MPII Multi-Person Pose Dataset. 6.2 Results We ﬁrst evaluate the impact of the parameter ˝in f ˝(s) (21) on the pose estimation accuracy measured as mean AP on the validation set containing 1200 image”
DeepPose: Human Pose Estimation via Deep Neural Networks (2013, methodology): “A different approach to capture higher-order relationship is through image-dependent PS models, which can be estimated via a global classiﬁer [26, 20, 18].”
Pose for Action - Action for Pose (2016, background): “Popular approaches for pose estimation [2], [5], [6], [18], [29], [38] adopt the pictorial structure (PS) model, which resembles the human skeleton and allows for efficient inference in case of tree structures [7], [8].”