Papers and topics for the misc reading group. Compiled by Daniel Huber on 3/25/02 based on suggestions from the group. I have tried to organize the papers into categories that can be covered in a week or two of discussion. Themes ====== Theory Illumination cones Embeddings tutorial Robust methods and heteroscedastic noise (Meer and co.) Vision problems Estimating BRDFs from images Tracking Human and articulated motion tracking Face detection and recognition Gait recognition High-level vision Specific approaches Parts-based recognition Accelerating recognition and detection Occlusion and clutter 3D vision Extracting layers from images Stereo Structure and motion Non-rigid 3D 3D reconstruction (from images) 3D range data Sensors Omni-cam Misc papers "Maybe" papers ============================================================= Papers Illumination cones (Owen) ------------------------- P. Belhumeur and D. Kriegman, "What Is the Set of Images of an Object Under All Possible Illumination Conditions?" Int. Journal of Computer Vision, 28(3), 1998, PP. 245?60. www.cs.cmu.edu/~owenc/ijcv98cones.pdf A conference version of this paper won the best paper award at CVPR '96. It gives a derivation of what illumination cones are and how you can estimate them from 3 example images. P. Belhumeur, A. Georghiades and D. Kriegman, "From Few to Many: Illumination Cone Models for Face Recognition Under Variable Lighting and Pose," IEEE Trans. PAMI, 2001, pp.643-660. www.cs.cmu.edu/~owenc/pami01cones.pdf An illumination cone only covers views of the object at one pose. So why not many different cones for many different poses. NinePoints of Light: Acquiring Subspaces for Face Recognition under Variable Lighting. Kuang-Chih Lee, Jeffery Ho, David Kriegman / IEEE Conf. on Computer Vision and Pattern Recognition, 2001, pp.519-526. www.cs.cmu.edu/~owenc/cvpr01cones.pdf Shows that if you set up 9 light sources and a camera, and snap 9 photos, you can use those training images to do recognition over a wide range of lighting changes. Ronen Basri & David Jacobs - Lambertian Reflectance and Linear Subspaces Embeddings (we need a tutorial) ------------------------------- NMF (non-negative matrix factorization) Li, Hou, Zhang, Cheng - Learning Spatially Localized, Parts-Based Representation (CVPR 2001) ICA (independent components analysis) Using an ICA representation of high-dimensional data for object recognition and classification, Bressan, Guillamet, Vitria ISA (independent subspace analysis) Stan Z. Li, X. Lv, and H. Zhang - View-Based Clustering of Object Appearances Based on Independent Subspace Analysis (ICCV 01) Sam T. Roweis, Lawrence K. Saul, Geoffrey E. Hinton - Global Coordination of Local Linear Models (NIPS 01) [also had a paper in Science (need reference)] Pietro Perona and Marzia Polito - Grouping and dimensionality reduction by locally linear embedding (NIPS 01) Dasgupta - Experiments with random projection (Uncertainty in AI 2000) - referenced by Shapiro CVPR paper *Multi-dimensional scaling - Kruskal (Martial has book) Other ideas: Lewiki (sp?), blind source separation (application of ICA) Robust methods and heteroscedastic noise ---------------------------------------- Bride and Meer - Registration via direct methods: a statistical approach (CVPR 01) Chen, Meer and Tyler - Robust regression for data with multiple structures (CVPR 01) Vision problems =============== Recovering BRDFs from images ---------------------------- What papers? Tracking -------- Toyama and Blake - Probabilistic Tracking in a Metric Space (ICCV 01) - best paper award Black, M. J. and Jepson, A - EigenTracking: Robust matching and tracking of articulated objects using a view-based representation (ICCV 98) Avidan - Support vector tracking (CVPR 01) Ying Wu, Thomas S. Huang - A Co-inference Approach to Robust Visual Tracking (ICCV 01) Human and articulated motion tracking ------------------------------------- DiFranco, Cham, and Rehg - Reconstruction of 3D figure motion from 2D correspondences (CVPR 2001) David Liebowitz and Stefan Carlsson - Uncalibrated Motion Capture Exploiting Articulated Structure Constraints (ICCV 2001) Ralf Plankers and Pascal Fua - Articulated Soft Objects for Video-based Body Modeling (ICCV 2001) Choo and Fleet - People Tracking Using Hybrid Monte Carlo Filtering (ICCV 2001) Zhao, Nevatia, and Lv - Segmentation and tracking of multiple humans in complex situations (CVPR 01) Other papers to consider: Brand and Hertzmann - Style Machines (SIGGRAPH 00) [motion synthesis] Face detection and recognition ------------------------------ Detecting faces in images: a survey - Kriegman (PAMI 2002) Batur and Hayes - Linear subspaces for illumination robust face recognition (CVPR 2001 II-296) Paul Viola, Michael Jones - Robust Real-Time Face Detection (ICCV 01) Gait recognition ---------------- * Bissacco, Chiuso, Ma, Soatto - Recognition of Human Gaits (where?) * Nixon - (which paper? http://www.ecs.soton.ac.uk/~msn/) - Caught bank robbers Other ideas: Bob Collins, U Maryland HID work (e.g. A. Kale, A.N. Rajagopalan, N. Cuntoor and V. Kruger - Human Identification Using Gait) High-level vision ----------------- Antonio Torralba Pawan Sinha - Statistical Context Priming for Object Detection (ICCV 01) Kobus Barnard, David Forsyth - Learning the Semantics of Words and Pictures Specific approaches =================== Parts-based recognition ----------------------- Ioffe and Forsyth - Human Tracking with Mixtures of Trees (ICCV 2001) Ioffe - Mixtures of trees for object recognition (CVPR 01) Heisele, Serre, Pontil, Poggio - Component-based Face Detection (CVPR 2001) Bernd Heisele, Thomas Serre, Massimiliano Pontil, Thomas Vetter and Tomaso Poggio - Categorization by Learning and Combining Object Parts (NIPS 01) Lowe - Local feature view clustering for 3d object recognition (CVPR 2001) Accelerating recognition and detection -------------------------------------- Heisele, Serre, Mukherjee, Poggio - Feature Reduction and Hierarchy of Classifiers for Fast Object Detection in Video Images [speed up object detection - not a parts-based recognition paper] Jones - Rapid Object Detection using a Boosted Cascade of Simple Features (CVPR 01) Paul Viola and Michael Jones - Fast and Robust Classification using Asymmetric AdaBoost and a Detector Cascade Learning with occlusion and clutter ----------------------------------- Ying and Castanon - Feature Based Object Recognition using Statistical Occlusion Models with One-to-one Correspondence (ICCV 2001) Selinger and Nelson - Minimally supervised acquisition of 3d recognition models from cluttered images (CVPR 01) 3D vision ========= Extracting layers from images ----------------------------- Jojic and Frey - Learning Flexible Sprites in Video Layers (CVPR 01) - separate video images into layers Kifa Ke and Kanade - A subspace approach to layer extraction Chuang, Curless, Salesin, and Szeliski - A bayesian approach to digital matting (CVPR 01) [there were several other papers on alpha blending - Tomasi?] Stereo ------ Gluckman and Nayar - Rectifying transformations that minimize resampling effects (CVPR 01) Kolmogorov and Zabih - Computing Visual Correspondence with Occlusions using Graph Cuts registration (ICCV 2001) Berg and Malik - Geometric blur for template matching (maybe) Structure and motion -------------------- Irani - Multi-frame optical flow estimation using subspace constraints (ICCV 99) Caspi and Irani - Alignment of non-overlapping sequences (ICCV 01) Schulz, Burgard, Fox, and Cremers - Tracking multiple moving objects with a mobile robot (CVPR 01) Fitzgibbon - Simultaneous linear estimation of multiple view geometry and lens distortion (CVPR 01) Non-rigid 3D ------------ Torresani, Yang, Alexander, and Bregler - Tracking and modeling non-rigid objects with rank constraints (CVPR 01) Brand and Bhotika - Flexible flow for 3D nonrigid tracking and shape recovery (CVPR 01) Brand - Morphable 3d models from video (CVPR 01) Rangarajan, Chui, Mjolsness - A relationship between spline-based deformable models and weighted graphs in non-rigid matching (CVPR 01) 3D reconstruction (from images) ------------------------------- Seth Teller et al - Calibrated, registered images of an extended urban area (CVPR 01) [Also, we should do a general overview of Teller's entire project.] Kawaski, Ikeuchi, and Sakauchi - Light field rendering for large-scale scenes (CVPR 01) Xiong Olson, and Matthies - Computing depth maps from descent imagery (CVPR 01) Zhang, Phocion, Samson, and Seitz - Single view modeling of free-form surfaces (CVPR 01) Bartoli - Piecewise planar segmentation for automatic scene modeling (CVPR 01) Agrawal and Davis - A probabilistic framework for surface reconstruction from multiple images (CVPR 01) [I think there is another paper on this?] 3D range data ------------- Sun and Abidi - Surface Matching by 3D Point Fingerprint (ICCV 01) Fruh and Zakhor - 3d model generation for cities using arial photographs and ground level laser scans (CVPR 01) Sagawa, Nishino, and Ikeuchi - Robust and adaptive integration of multiple range images with photometric attributes (CVPR 01) [also Nishino ACCV paper] Sensors ======= Omnicam ------- Geyer and Daniilidis - Structure and motion from uncalibrated catadioptric views (CVPR 01) Hicks and Perline - Geometric distributions for catadioptric sensor design (CVPR 01) Misc ---- Olivier Chapelle, Bernhard Scholkopf - Incorporating Invariances in Nonlinear Support Vector Machines (NIPS 01) [theory?] Jojic, Simard, Frey, and Heckerman - Separating Appearance from Deformation (ICCV 01) [recognition] Serge Belongie, Jitendra Malik and Jan Puzicha - Matching Shapes (ICCV 01) Keren - Anti-Sequences: Event Detection by Frame Stacking (CVPR 01) Brendan J. Frey, Nebojsa Jojic - Fast, large-scale transformation-invariant clustering (NIPS 01) Hadjidemetriou, Grossberg, and Nayar - Spatial information in multiresolution histograms (CVPR 01) [theory?] Dror, Leung, Adelson, and Willsky - Statistics of real-world illumination (CVPR 01) Ballester, V. Caselles, J. Verdera, M. Bertalmio and Sapiro - A Variational Model for Filling-In Gray Level and Color Images (ICCV 01) - image processing Soatto, Doretto, Wu - Dynamic textures (ICCV 01) - texture of moving objects like waves Maybe papers ------------ Superresolution --------------- Liu, Shum, Zhang. - A two step approach to hallucinating faces (CVPR 01) D. Capel and A. Zisserman - Super-Resolution From Multiple Views Using Learnt Image Models (CVPR 01) Learning -------- Chawla et al - Bagging Is A Small-Data-Set Phenomenon (CVPR 01) CH Li - Constrained Minimum Cut For Classification Using Labeled and Unlabeled Data (CVPR 01) Felzenszwalb - Learning models for object recognition (CVPR 01) Schechner - Instant Dehazing of Images Using Polarization (CVPR 01) ---------------------------------------- Ideas from CVPR 2003: Face Recognition Under Variable Lighting using Harmonic Image Exemplars L. Zhang, D. Samaras (SUNY@Stony Brook) - the idea was to extrac harmonic image exemplars from 3D models and use that to do lighting invariant recognition from a single example - also, we should look at papers leading up to this. General topic: Harmonic images A Variational Framework for Image Segmentation Combining Motion Estimation and Shape Regularization D. Cremers (University of California) - There is a similar ECCV paper that we should also discuss at the same time Kinematic Jump Processes For Monocular 3D Human Tracking C. Sminchisescu, W. Triggs (INRIA) - Danny will look at Statistics of Shape via Principal Component Analysis on Lie Groups P. T. Fletcher, C. Lu, S. Joshi (University of North Carolina) - instead of principal componental analysis, principal geodesic (curve) analysis - some interesting results were shown with a liver model Representation and detection of deformable shapes P. Felzenszwalb (Massachusetts Institute of Technology) - the representation uses delaney triangulation - Also look at related papers by Perona A New Graph-theoretic Approach to Clustering and Segmentation M. Pavan, M. Pelillo (University of Venice) - Questionable -- Sanjiv will look at. Optimal Linear Representations of Images for Object Recognition X. Liu, A. Srivastava, K. Gallivan (Florida State University) - the idea was to directly optimize a linear transform for discrimination Error Analysis of 3D Motion Estimation Algorithms in the Differential Case T. Xiang (University of London) L.-F. Cheong (National University of Singapore) - Sanjiv will look at. A Novel Support Vector Classifier with Better Rejection Performance C. Yuan, D. Casasent (Carnegie Mellon University) - should we have them come and present? An Efficient Approach to Learning Inhomogeneous Gibbs Model Z. Liu, H. Chen, H. Shum (Microsoft Research) - Sanjiv will discuss together with KL boosting paper below - Also combine with other papers on feature selection (e.g., Nuno Vasconcelos below) Feature selection by maximum marginal diversity: optimality and implications for visual recognition N. Vasconcelos (HP Cambridge Research Laboratory) - is there something interesting here or is it really just mutual information - or did he prove something interesting here about using mutual information Kullback-Leibler Boosting and Its Application to Face Detection C. Liu, H. Shum (Microsoft Research) Structure from motion for scenes without features A. Yezzi (Georgia Institute of Technology) S. Soatto (University of California, Los Angeles) Combine next two papers and discuss Helmholtz recip constraint. Surface Reconstruction via Helmholtz Reciprocity with a Single Image Pair P. Tu, P. Mendonca (GE Global Research Center) Toward a Stratification of Helmholtz Stereopsis T. Zickler (Yale University) P. Belhumeur (Columbia University) D. Kriegman (University of California, San Diego) Independent Component Analysis in a Facial Local Residue Space T.-K. Kim, , H. Kim, W. Hwang, S.-C. Kee (Samsung Advanced Institute of Technology) J. Kittler (University of Surrey) - Sanjiv will look at Nonparametric Belief Propagation E. Sudderth, A. Ihler, W. Freeman, A. Willsky (Massachusetts Institute of Technology) - Combine with next paper (very similar) Pampas: Real-Valued Graphical Models for Computer Vision M. Isard (Microsoft Research) Generalized Principal Component Analysis R. Vidal (University of California, Berkeley) Y. Ma (University of Illinois) S. Sastry (University of California, Berkeley) - the math seemed interesting - generating PCA via polynomial solving - but how reliant / constrained is the solution because of the assumption of no noise - Martial - nice idea, but see old paper by Cooper on fitting polynomials - they are unstable under noisy conditions. Optimal Segmentation of Dynamic Scenes from Feature Points or Image Intensities R. Vidal, S. Sastry (University of California, Berkeley) - App of Generalized PCA? Combine with previous paper Vector-Valued Image Regularization with PDE's: A Common Framework for Different Applications D. Tschumperle, R. Deriche (INRIA) - results seem interesting - let's look at general topic of PDEs in vision - e.g., Faugeras' group Wide-baseline multiple-view correspondences V. Ferrari (ETH Zurich) T. Tuytelaars (KUL) L. Van Gool (KUL/ETHZ) - Danny will look at Local Appearance-Based Models using High-Order Statistics of Image Features B. Moghaddam (Mitsubishi Electric Research Laboratories) D. Guillamet, J. Vitria (Computer Vision Center (CVC)) - some version of ICA 3D Surface Modeling from Curves D. Tubic, P. Hebert, D. Laurendeau (Laval University) - Danny will look at. Object Segmentation Using Graph Cuts Based Active Contours N. Xu (University of Illinois at Urbana-Champaign) R. Bansal (Columbia University) N. Ahuja (University of Illinois at Urbana-Champaign) Constrained Subspace Modelling J. Vermaak (Cambridge University) P. Perez (Microsoft Research) - the idea here was to introduce a prior to constrain the subspace Practical Non-parametric Density Estimation on a Transformation Group for Vision E. Miller (University of California, Berkeley) C. Chefd'hotel (INRIA) Constraint on Five Points in Two Images T. Werner (Oxford Univeresity) - honorable mention paper Object class recognition by unsupervised scale-invariant learning R. Fergus (Oxford University & California Institute of Technology) P. Perona (California Institute of Technology) A. Zisserman (Oxford University) - best paper award - similar to previous perona paper. - also look at iccv - things different from old Perona stuff - new interest operator - intrinsic scale - patch appearance model - large - possible issue - when trained on 0% faces still get 10% right, this probably means that the some test images had very little clutter 3D Object Modeling and Recognition Using Affine-Invariant Patches and Multi-View Spatial Constraints F. Rothganger, S. Lazebnik (University of Illinois, Urbana-Champaign) C. Schmid (INRIA) J. Ponce (University of Illinois, Urbana-Champaign) - this seemed pretty interesting to me - but they did have to use a highly textured teddy bear object - the extra affine alignment step also seemed interesting Properties and applications of shape recipes A. Torralba, W. Freeman (Massachusetts Institute of Technology) Space-time stereo? Do we want to cover this? Space-Time Stereo: A Unifying Framework for Depth from Triangulation J. Davis (Honda Fundamental Research Laboratories) R. Ramamoorthi (Columbia University) S. Rusinkiewicz (Princeton University) Spacetime Stereo: Shape recovery for dynamic scenes L. Zhang, B. Curless, S. Seitz (University of Washington) Estimating 3D Hand Pose from a Cluttered Image V. Athitsos, S. Sclaroff (Boston University) Enhancing Image and Video Retrieval: Learning via Equivalence Constraints T. Hertz, N. Shental, A. Bar-Hillel, D. Weinshall (Hebrew University) Texture Classification: Are Filter Banks Necessary? M. Varma, A. Zisserman (University of Oxford) Object Removal by Exemplar-Based Inpainting A. Criminisi, P. Perez, K. Toyama (Microsoft Research) - results look good Image Hallucination with Primal Sketch Priors J. Sun (Microsoft Research) N.-N. Zhang (Xian Jiaotong University) H. Tao (Univ. of California, Santa Cruz) H. Shum (Microsoft Research)