The Shape Boltzmann Machine: A Strong Model of Object Shape

Eslami, S. M. Ali; Heess, Nicolas; Williams, Christopher K. I.; Winn, John

doi:10.1007/s11263-013-0669-1

Cited by 110 publications

(87 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GPLVM [17] GPLVMDT [22] DBN [13] SBM [10] VAE [15] ∼ InfoGAN [5] ∼ ShapeOdds [8] This work Table 1. Summary of properties of related models.…”

Section: Non-gaussian Likelihood Explicit Smooth Low-dim Manifold Fulmentioning

confidence: 99%

“…In our work we focus on recent unsupervised statistical models that operate directly on the pixel domain. Interest in these models was revived by the Shape Boltzmann Machine (SBM) work of Eslami et al [10] and they have been shown to be useful for a variety of vision applications [9,16,29]. These deep models can also be readily extended into the 3D domain, e.g., by recent work on 3D ShapeNets [32].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Gaussian Process Deep Belief Networks: A Smooth Generative Model of Shape with Uncertainty Propagation

Martino

Bodin

et al. 2019

Computer Vision – ACCV 2018

View full text Add to dashboard Cite

The shape of an object is an important characteristic for many vision problems such as segmentation, detection and tracking. Being independent of appearance, it is possible to generalize to a large range of objects from only small amounts of data. However, shapes represented as silhouette images are challenging to model due to complicated likelihood functions leading to intractable posteriors. In this paper we present a generative model of shapes which provides a low dimensional latent encoding which importantly resides on a smooth manifold with respect to the silhouette images. The proposed model propagates uncertainty in a principled manner allowing it to learn from small amounts of data and providing predictions with associated uncertainty. We provide experiments that show how our proposed model provides favorable quantitative results compared with the state-of-the-art while simultaneously providing a representation that resides on a low-dimensional interpretable manifold.

show abstract

“…GPLVM [17] GPLVMDT [22] DBN [13] SBM [10] VAE [15] ∼ InfoGAN [5] ∼ ShapeOdds [8] This work Table 1. Summary of properties of related models.…”

Section: Non-gaussian Likelihood Explicit Smooth Low-dim Manifold Fulmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Gaussian Process Deep Belief Networks: A Smooth Generative Model of Shape with Uncertainty Propagation

Martino

Bodin

et al. 2019

Computer Vision – ACCV 2018

View full text Add to dashboard Cite

show abstract

“…Other DGMs include deep Boltzmann machines (DBMs) [SH09]. DBMs are undirected models that have been used to capture complex distributions over speech data [MDH12], images [EHWW14, RHSW11] and part‐segmented objects [HKM15]. Although DBMs are flexible, they can be difficult and time‐consuming to train, and are more complicated to sample from than directed models.…”

Section: Related Workmentioning

confidence: 99%

The shape variational autoencoder: A deep generative model of part‐segmented 3D objects

Nash

Williams

2017

Computer Graphics Forum

136

View full text Add to dashboard Cite

We introduce a generative model of part‐segmented 3D objects: the shape variational auto‐encoder (ShapeVAE). The ShapeVAE describes a joint distribution over the existence of object parts, the locations of a dense set of surface points, and over surface normals associated with these points. Our model makes use of a deep encoder‐decoder architecture that leverages the part‐decomposability of 3D objects to embed high‐dimensional shape representations and sample novel instances. Given an input collection of part‐segmented objects with dense point correspondences the ShapeVAE is capable of synthesizing novel, realistic shapes, and by performing conditional inference enables imputation of missing parts or surface normals. In addition, by generating both points and surface normals, our model allows for the use of powerful surface‐reconstruction methods for mesh synthesis. We provide a quantitative evaluation of the ShapeVAE on shape‐completion and test‐set log‐likelihood tasks and demonstrate that the model performs favourably against strong baselines. We demonstrate qualitatively that the ShapeVAE produces plausible shape samples, and that it captures a semantically meaningful shape‐embedding. In addition we show that the ShapeVAE facilitates mesh reconstruction by sampling consistent surface normals.

show abstract

“…Eslami and Williams [10] extend the Shape Boltzmann Machine [17] model (SBM), applied for the task of modeling binary shape images, to account for the object's parts. Their Multinomial SBM is combined with an appearance model to form a fully generative model of images of objects.…”

Section: Related Workmentioning

confidence: 99%

Shape-Based Pedestrian Segmentation in Still Images

Jacques

Musse

2016

Int. J. Semantic Computing

View full text Add to dashboard Cite

Pedestrian segmentation is a problem of considerable practical interest. In this work we present an extended version of our shape-based model for pedestrian segmentation, which can also be used to give an initial guess of the 2D pedestrians pose/orientation. The proposed model is initialized by a bounding-box of the person under analysis, which can be estimated by a person detector. The basic idea of the proposed model is to create a graph around the detected person, based on a scale invariant shape model and the estimated contour is given by a path in the graph that maximizes certain boundary energy. In practice, such energy should be large in the boundary between the foreground/background. To cope with pose/shape variations, the¯nal estimate is given by a selection scheme, which takes into consideration the individual estimate given by di®erent generated graphs. Experimental results indicated that the proposed technique works well in non trivial images, with comparable accuracy to the state-of-the-art. Int. J. Semantic Computing 2016.10:53-71. Downloaded from www.worldscientific.com by MONASH UNIVERSITY on 06/11/16. For personal use only.The pedestrian segmentation task can be considered a special case of person/ human segmentation. As related in the work of Dollar et al.[2], people analysis, by computer vision techniques, makes the use of datasets containing people in unconstrained pose in a wide range of domains whilst the area of pedestrian analysis uses datasets containing upright people (standing or walking), typically viewed from more restricted viewpoints.Pedestrian segmentation can be used in several applications, including robotics, surveillance systems, driver assistance models, among others. Farenzena and his group [3], for example, assume the presence of the silhouette of an individual, obtained for each person by inferring over the STEL generative model [4], to extract appearance features applied in a person re-identi¯cation problem. Similarly, Ma et al.[5] adopted a pedestrian parsing (partitioning the human body into semantic regions, such as head, torso, arms, etc, which may include clothes and accessories) model [6] to extract appearance features of speci¯c pedestrian's body parts for the same purpose (person re-identi¯cation). This trend seems to be very useful in human analysis (speci¯cally in person re-identi¯cation tasks), as the background usually include several noise, making even more di±cult to obtain a good solution for such task.In this work we extend our shape-based model for pedestrian segmentation in still images [7] to estimate the main orientation of pedestrian¯gures. In addition, other features of the model were evaluated, as considering grayscale images (instead of color¯gures) as well as the inclusion of a post-processing operation, which takes into consideration color cues. This journal version also includes new experimental results and a more detailed discussion of some technical points.As in the original work [7], we focus on the case where an external pedestrian detector i...

show abstract

The Shape Boltzmann Machine: A Strong Model of Object Shape

Cited by 110 publications

References 59 publications

Gaussian Process Deep Belief Networks: A Smooth Generative Model of Shape with Uncertainty Propagation

Gaussian Process Deep Belief Networks: A Smooth Generative Model of Shape with Uncertainty Propagation

The shape variational autoencoder: A deep generative model of part‐segmented 3D objects

Shape-Based Pedestrian Segmentation in Still Images

Contact Info

Product

Resources

About