Shape2Pose: Human Centric Shape Analysis CMPT888 Vladimir G. Kim Siddhartha Chaudhuri Leonidas Guibas Thomas Funkhouser Stanford University Princeton University.

Slides:



Advertisements
Similar presentations
Applications of one-class classification
Advertisements

Face Alignment by Explicit Shape Regression
Interactively Co-segmentating Topically Related Images with Intelligent Scribble Guidance Dhruv Batra, Carnegie Mellon University Adarsh Kowdle, Cornell.
A CTION R ECOGNITION FROM V IDEO U SING F EATURE C OVARIANCE M ATRICES Kai Guo, Prakash Ishwar, Senior Member, IEEE, and Janusz Konrad, Fellow, IEEE.
Support Vector Machines
Face Alignment with Part-Based Modeling
Patch to the Future: Unsupervised Visual Prediction
SA-1 Robotic Self-Perception and Body Scheme Learning Jürgen Sturm Christian Plagemann Wolfram Burgard University of Freiburg Germany.
Parsing Clothing in Fashion Photographs
Model Assessment, Selection and Averaging
Modeling the Shape of People from 3D Range Scans
Cambridge, Massachusetts Pose Estimation in Heavy Clutter using a Multi-Flash Camera Ming-Yu Liu, Oncel Tuzel, Ashok Veeraraghavan, Rama Chellappa, Amit.
Model base human pose tracking. Papers Real-Time Human Pose Tracking from Range Data Simultaneous Shape and Pose Adaption of Articulated Models using.
Real-Time Human Pose Recognition in Parts from Single Depth Images Presented by: Mohammad A. Gowayyed.
Proportion Priors for Image Sequence Segmentation Claudia Nieuwenhuis, etc. ICCV 2013 Oral.
Chapter 4: Linear Models for Classification
4/15/2017 Using Gaussian Process Regression for Efficient Motion Planning in Environments with Deformable Objects Barbara Frank, Cyrill Stachniss, Nichola.
Basis Expansion and Regularization Presenter: Hongliang Fei Brian Quanz Brian Quanz Date: July 03, 2008.
Watching Unlabeled Video Helps Learn New Human Actions from Very Few Labeled Snapshots Chao-Yeh Chen and Kristen Grauman University of Texas at Austin.
Instructor: Mircea Nicolescu Lecture 13 CS 485 / 685 Computer Vision.
A Study of Approaches for Object Recognition
Predictive Automatic Relevance Determination by Expectation Propagation Yuan (Alan) Qi Thomas P. Minka Rosalind W. Picard Zoubin Ghahramani.
Curve Analogies Aaron Hertzmann Nuria Oliver Brain Curless Steven M. Seitz University of Washington Microsoft Research Thirteenth Eurographics.
Machine Learning CUNY Graduate Center Lecture 3: Linear Regression.
Discriminative Training of Kalman Filters P. Abbeel, A. Coates, M
P. Rodríguez, R. Dosil, X. M. Pardo, V. Leborán Grupo de Visión Artificial Departamento de Electrónica e Computación Universidade de Santiago de Compostela.
© 2013 IBM Corporation Efficient Multi-stage Image Classification for Mobile Sensing in Urban Environments Presented by Shashank Mujumdar IBM Research,
Radial Basis Function Networks
Machine Learning in Simulation-Based Analysis 1 Li-C. Wang, Malgorzata Marek-Sadowska University of California, Santa Barbara.
Face Alignment Using Cascaded Boosted Regression Active Shape Models
IMPLEMENTATION ISSUES REGARDING A 3D ROBOT – BASED LASER SCANNING SYSTEM Theodor Borangiu, Anamaria Dogar, Alexandru Dumitrache University Politehnica.
Extracting Places and Activities from GPS Traces Using Hierarchical Conditional Random Fields Yong-Joong Kim Dept. of Computer Science Yonsei.
Machine Learning CUNY Graduate Center Lecture 3: Linear Regression.
Constraints-based Motion Planning for an Automatic, Flexible Laser Scanning Robotized Platform Th. Borangiu, A. Dogar, A. Dumitrache University Politehnica.
Parameter selection in prostate IMRT Renzhi Lu, Richard J. Radke 1, Andrew Jackson 2 Rensselaer Polytechnic Institute 1,Memorial Sloan-Kettering Cancer.
1 Energy-aware stage illumination. Written by: Friedrich Eisenbrand Stefan Funke Andreas Karrenbauer Domagoj Matijevic Presented By: Yossi Maimon.
Multimodal Interaction Dr. Mike Spann
Niloy J. Mitra Leonidas J. Guibas Mark Pauly TU Vienna Stanford University ETH Zurich SIGGRAPH 2007.
1 TEMPLATE MATCHING  The Goal: Given a set of reference patterns known as TEMPLATES, find to which one an unknown pattern matches best. That is, each.
Segmental Hidden Markov Models with Random Effects for Waveform Modeling Author: Seyoung Kim & Padhraic Smyth Presentor: Lu Ren.
NATIONAL TECHNICAL UNIVERSITY OF ATHENS Image, Video And Multimedia Systems Laboratory Background
ALIGNMENT OF 3D ARTICULATE SHAPES. Articulated registration Input: Two or more 3d point clouds (possibly with connectivity information) of an articulated.
COMMON EVALUATION FINAL PROJECT Vira Oleksyuk ECE 8110: Introduction to machine Learning and Pattern Recognition.
Yao, B., and Fei-fei, L. IEEE Transactions on PAMI(2012)
Enforcing Constraints for Human Body Tracking David Demirdjian Artificial Intelligence Laboratory, MIT.
Automatic Image Annotation by Using Concept-Sensitive Salient Objects for Image Content Representation Jianping Fan, Yuli Gao, Hangzai Luo, Guangyou Xu.
MURI: Integrated Fusion, Performance Prediction, and Sensor Management for Automatic Target Exploitation 1 Dynamic Sensor Resource Management for ATE MURI.
Exploiting Context Analysis for Combining Multiple Entity Resolution Systems -Ramu Bandaru Zhaoqi Chen Dmitri V.kalashnikov Sharad Mehrotra.
Vision-based human motion analysis: An overview Computer Vision and Image Understanding(2007)
Stable Multi-Target Tracking in Real-Time Surveillance Video
D. M. J. Tax and R. P. W. Duin. Presented by Mihajlo Grbovic Support Vector Data Description.
Interactive Learning of the Acoustic Properties of Objects by a Robot
MURI Annual Review, Vanderbilt, Sep 8 th, 2009 Heterogeneous Sensor Webs for Automated Target Recognition and Tracking in Urban Terrain (W911NF )
Chapter 7. Learning through Imitation and Exploration: Towards Humanoid Robots that Learn from Humans in Creating Brain-like Intelligence. Course: Robots.
Sponsored by Deformation-Driven Topology-Varying 3D Shape Correspondence Ibraheem Alhashim Kai Xu Yixin Zhuang Junjie Cao Patricio Simari Hao Zhang Presenter:
A Multiresolution Symbolic Representation of Time Series Vasileios Megalooikonomou Qiang Wang Guo Li Christos Faloutsos Presented by Rui Li.
Predicting Short-Term Interests Using Activity-Based Search Context CIKM’10 Advisor: Jia Ling, Koh Speaker: Yu Cheng, Hsieh.
RiskTeam/ Zürich, 6 July 1998 Andreas S. Weigend, Data Mining Group, Information Systems Department, Stern School of Business, NYU 2: 1 Nonlinear Models.
Ali Ghadirzadeh, Atsuto Maki, Mårten Björkman Sept 28- Oct Hamburg Germany Presented by Jen-Fang Chang 1.
San Diego May 22, 2013 Giovanni Saponaro Giampiero Salvi
Semi-Global Matching with self-adjusting penalties
Real-Time Human Pose Recognition in Parts from Single Depth Image
Dynamical Statistical Shape Priors for Level Set Based Tracking
Synthesis of Motion from Simple Animations
Introduction to Radial Basis Function Networks
Dimitris Valeris Thijs Ratsma
Machine Learning – a Probabilistic Perspective
One-shot learning and generation of dexterous grasps
Point Set Representation for Object Detection and Beyond
Presentation transcript:

Shape2Pose: Human Centric Shape Analysis CMPT888 Vladimir G. Kim Siddhartha Chaudhuri Leonidas Guibas Thomas Funkhouser Stanford University Princeton University Paper Study Presentation by Shunan

Content A fundamental question: “how do people interact with objects?” 1.Problem Introduction. 2.Major Problems. 3.Proposed Method. 4.Conclusion & Discussion.

1 Problem introduction

The World need 3D models. Methods for saliency estimation, shape segmentation, feature detection, symmetry analysis, and surface correspondence; These method mainly using techniques that compute global shape properties. Object affordance – a quality of an object that allows someone to perform an action. Main technical contribution is a polynomial-time optimization algorithm that finds an approximate minimum in the combinatorial space of body-to contact assignments. Main challenge in affordance for shape analysis is to automatically predict the pose that a human will take when using a given object.

Main contributions 1.Shape2Pose, a novel affordance-inspired shape analysis tool. 2.Polynomial-time algorithm for exploring the combinatorial space. 3.A dataset of 147 models from diverse object classes with annotated ground truth poses. 4.Evaluate our method and demonstrate favorable performance as results. Shape2Pose Polynomial Datasets Results

2 Major Problems

Major Problems 01 Extend pervious toolkit with affordance analysis that predicts contact points and pose parameters, essentially detecting invariant structural relations. Shape Analysis. 02 Do not assume that there is a small set of discrete poses; instead, Search a continuous pose parameter space. This finer representation enables higher accuracy and applications beyond shape classification. Affordance. 03 Previous approaches focus on robot and virtual hand interaction, while we focus on shape analysis. Also note that a grasping interaction has a simple functional purpose. Grasp Prediction.

Shape2Pose Shape2Pose: Simultaneously predicts kinematic parameters for a static human pose and points of contact between a human body and the shape’s. Main difference: Extra critical information, the shape and deformation modes of a human body. “What significant patterns you can find in this shape?” VS “How can this human body be posed to interact with this shape?”

2 Find Pattern in Shapes. Interact with shapes.

3 Proposed Method

System pipeline Shap2epose Training Affordance Model Contact Distance Feature Compatibility. Pose Prior Symmetry & Surface Intersection Predicting Pose Model Match Inverse Kinematics

Affordance model is defined as the minimizer of an energy function above Training Affordance Model The goal in this stage is to build an energy function that can be used to evaluate the interaction between a shape S and a human pose represented by a rigid transformation T. w dist = 1000, w feat = 10, w pose = 0.3, w sym = 1, w isect = 0.05 For all.

E dist(T, Ɵ,m,S): Contact Distance High weight : parts do not touch the target points. If a body part is assigned to a surface point, we want to ensure they actually make contact. If not, we penalize. Ɵ : Joint angles m: Key body parts with constraints S: Object Shape T: Human pose transformation

E feat(m,S): Feature Compatibility High weight : if local surface is ill-suited for placement. Which means hands to graspable, butt on seat. Give a set of training shape: {S 1, S 2 …S m } with ground truth, Regression Ɵ: Joint angles m: Key body parts with constraints S: Object Shape T: Human pose transformation

E pose( Ɵ ): Pose Prior. Distinguishes unreal pose. learn the pose prior from training examples to get the distribution. Ɵ: Joint angles m: Key body parts with constraints S: Object Shape T: Human pose transformation

E sym(T,m,S): Symmetry. Observe parts of shapes with which humans interact typically have local bilateral symmetry. Ɵ: Joint angles m: Key body parts with constraints S: Object Shape T: Human pose transformation E isect(T, Ɵ,S) : Surface Intersection Helps avoid intersections between the shape and the human. Thickness of body mainly.

For training Input : A collection of shapes with manually prescribed contact points. Output: An affordance model represented by an objective function that measures the quality of a pose for any shape. Predicting a Human Pose Input : A novel shape. Output: A set of joint angles and contact points for a likely human interaction pose. For Predicting 2

Inverse Kinematics Predicting 1.if contact points m are assigned, one can solve for (T, Ɵ ) using inverse kinematics. 2.given T and Ɵ, one can infer contact points m via nearest neighbors. The key: possible to sample high-probability contact assignments m and high-probability poses Ɵ independently. They contribute to different major energy terms E feat (m) and E pose ( Ɵ ) respectively. M: key body part and its constraints Ɵ : Joints angles.

4 Conclusion & Discussion.

1.Predicting shape affordances is useful for computer analysis of 3D models. 2.Implementing a novel algorithm for generating the static pose. 3.Our algorithm uses a combination of local anthropometric classifiers as well as global biomechanics constraints. 4.Algorithm produces results within a 20cm tolerance in the vast majority of cases. Shape2Pose for analysis of 3D models Conclusion

Several limitations 1.Only static analysis of an object’s shape, not sufficient to understand all functional interactions with a human. 2.Discussed only a few low-level shape analysis applications in this paper. 3.Highly rely on training data.

CMPT 888 Thank You By Shunan