Strongly coupled problems - Computer Graphics at Stanford University
Transcription
Strongly coupled problems - Computer Graphics at Stanford University
http://www.cs.tau.ac.il/~wolf/OR/img/street_annotated.jpg CS 148, Summer 2012 Introduction to Computer Graphics and Imaging Justin Solomon Final exam Saturday 8/18/12, 12:15pm-3:15pm Makeup: Thursday 8/16/12, 1pm-4pm Two double-sided 8.5x11 sheets of notes Homework 6 Due Tuesday, August 14, 11:59pm Must be returned by Friday, August 17 ILM's VFX Pipeline and the Future of Performance Capture Hao Li Industrial Light and Magic www.youtube.com/watch?v=UKKv2fhX2TM Graphics produces images. Vision analyzes images. Inverse problems Graphics produces images. Vision analyzes images. Inverse problems Strongly coupled problems http://delivery.acm.org/10.1145/1110000/1103923/cs22.pdf?ip=171.67.216.21&acc=ACTIVE%20SERVICE&CFID=102787391&CFTOKEN=32277993&__acm__=1344483871_d2aba55345e66ac180cfb4b589ddb7ad Strongly coupled problems http://www.engineeringspecifier.com/public/primages/pr1200.jpg http://twr.cs.kuleuven.be/images/pointCloudProcessing.jpg http://www.ibe.kagoshima-u.ac.jp/~cgv/research/MVS.html Strongly coupled problems http://research.microsoft.com/en-us/um/people/jiansun/papers/dehaze_cvpr2009.pdf Strongly coupled problems http://graphics.cs.cmu.edu/projects/scene-completion/scene-completion.pdf Strongly coupled problems http://www.mpi-inf.mpg.de/~thormae/paper/CVPR11.pdf Strongly coupled problems http://www.youtube.com/watch?v=hnP7G7ahuus Strongly coupled problems http://www.youtube.com/watch?v=OmTCxff-DSk Strongly coupled problems http://graphics.cs.cmu.edu/projects/imageshaving/nguyen_eurographics_08. Strongly coupled problems http://graphics.cs.cmu.edu/projects/imageshaving/nguyen_eurographics_08. Strongly coupled problems Meaningful RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB RGB Not meaningful http://upload.wikimedia.org/wikipedia/commons/1/16/Cactus_flower_unidentified.jpg http://www.visitingdc.com/images/eiffel-tower-picture-2.jpg http://wondrouspics.com/wp-content/uploads/2011/10/eiffel-tower.jpg Alignment http://www.visitingdc.com/images/eiffel-tower-picture.jpg http://www.visitingdc.com/images/eiffel-tower-at-night.jpg http://www.hdwallpapers.in/walls/eiffel_tower_at_night_paris_francenormal.jpg Lighting and materials http://a4.ec-images.myspacecdn.com/images01/24/69bf8c4862e568739d8e282570465059/l.jpg http://cache.daylife.com/imageserve/0eRdbnFger62Z/439x.jpg http://gallery.weddingbee.com/photo/eloped-getting-married-in-front-of-the-eiffel-tower-in-paris http://specialtysites.typepad.com/.a/6a01127970a11f28a40120a5e8924b970b-320wi Occlusion http://ars.sciencedirect.com/content/image/1-s2.0-S0097849305000464-gr15.jpg http://graphics.cs.cmu.edu/projects/stvk/ Deformation http://upload.wikimedia.org/wikipedia/commons/thumb/5/53/Maurice_koechlin_pylone.jpg/220px-Maurice_koechlin_pylone.jpg http://www.papertoys.com/images/eiffel.gif http://farm7.static.flickr.com/6045/6230445674_7259927bcf.jpg http://cdn1.retronaut.co/wp-content/uploads/2010/07/Eiffel-Tower-6.jpg http://www.visitingdc.com/images/eiffel-tower-las-vegas.jpg https://s3.amazonaws.com/luuux-original-files/bookmarklet_uploaded/eiffel-tower-2.jpg Instances http://vision.stanford.edu/teaching/cs231a/lecture/lecture%2019_advanced_topics_cs231a.pdf http://vision.stanford.edu/teaching/cs231a/lecture/lecture%2019_advanced_topics_cs231a.pdf http://vision.stanford.edu/teaching/cs231a/lecture/lecture%2019_advanced_topics_cs231a.pdf http://vision.stanford.edu/teaching/cs231a/lecture/lecture%2019_advanced_topics_cs231a.pdf Computer graphics is a humongous field. Computer vision is a ^ humongous field. Provide examples of vision techniques applied in graphics. http://vision.middlebury.edu/flow/floweval-ijcv2011.pdf Image alignment http://digital-photography-school.com/wp-content/uploads/2009/03/exposure-fusion1.jpg Fuse exposures to one floating-point image http://research.microsoft.com/en-us/um/redmond/projects/flashnoflash/flash_no_flash%20%28web%29.pdf http://vision.in.tum.de/research/optical_flow_estimation Optical flow http://vision.in.tum.de/research/optical_flow_estimation Optical flow Flat texture gives no motion cues! http://upload.wikimedia.org/wikipedia/commons/f/f0/Aperture_problem_animated.gif http://2.bp.blogspot.com/-6LnK6MaXArg/TiLPPgmmyMI/AAAAAAAABdU/b87U2CDESfo/s1600/barber-shop-pole.jpg Stanford CS448F, Andrew Adams Global alignment Rotate Zoom Translate code.ucsd.edu/pcosman/253video1.ppt Rotate Zoom Translate code.ucsd.edu/pcosman/253video1.ppt p ~ 7! M~ p + ~t Affine Similarity Rigid http://faculty.cs.tamu.edu/schaefer/research/mls.pdf p~1 7! ~q1 p~2 7! ~q2 p~3 7! ~q3 ¢ ¢ ¢ 7! ¢ ¢ ¢ p~n 7! ~qn Stanford CS448F, Andrew Adams Example points p~1 7! ~q1 p~2 7! ~q2 p~3 7! ~q3 ¢ ¢ ¢ 7! ¢ ¢ ¢ p~n 7! ~qn Two differences: 1.Might make mistakes matching handles 2.Global motion Example points Flat Edge Corner No change in all directions No change along the edge direction Significant change in all directions MIT 6.882, Bill Freeman Harris corner detector E(¢x; ¢y) = X x;y w(x; y) [I(x + ¢x; y + ¢y) ¡ I(x; y)] Autocorrelation: How well shifting by (Δx, Δy) preserves image in area covered by w. w(x; y) = or MIT 6.882, Bill Freeman Harris corner detector 2 E(¢x; ¢y) = X x;y w(x; y) [I(x + ¢x; y + ¢y) ¡ I(x; y)] @I @I I(x + ¢x; y + ¢y) ¼ I(x; y) + ¢x + ¢y @x @y E(¢x; ¢y) ¼ M= ¡ ¢x ¢y X x;y w(x; y) ¢ µ M µ Ix2 Ix Iy ¢x ¢y ¶ Ix Iy Iy2 ; ¶ Harris corner detector 2 2 2 R = det M ¡ ·(tr M) = ¸1¸2 ¡ ·(¸1 + ¸2) E(¢x; ¢y) ¼ M= ¡ ¢x ¢y X x;y w(x; y) ¢ µ M µ Ix2 Ix Iy ¢x ¢y ¶ Ix Iy Iy2 ; ¶ Harris corner detector R MIT 6.882, Bill Freeman Harris corner detector Threshold MIT 6.882, Bill Freeman Harris corner detector Local max MIT 6.882, Bill Freeman Harris corner detector Features MIT 6.882, Bill Freeman Harris corner detector Features MIT 6.882, Bill Freeman Harris corner detector Features MIT 6.882, Bill Freeman Harris corner detector Match points Find transformation while ignoring outliers Rotation invariance Compute relative to gradient direction rI = (Ix; Iy ) Scale invariance Rescale w and its analogs multiple times MIT 6.882, Bill Freeman http://media.wiley.com/wires/WICS2.2/mfig001.jpg Match descriptors Invariant to: • Rotation • Scale • Intensity change • (Some) affine motion http://ryanlei.files.wordpress.com/2011/03/sift_descriptor2.jpg Scale-Invariant Feature Transform Invariant to: • Rotation • Scale • Intensity change • (Some) affine motion http://ryanlei.files.wordpress.com/2011/03/sift_descriptor2.jpg Scale-Invariant Feature Transform RANSAC: Random Sample Consensus Repeat: 1.Guess minimum number of points to determine parameters 2.Check if model works for other points http://upload.wikimedia.org/wikipedia/commons/d/de/Fitted_line.svg Random sampling MIT 6.882, Bill Freeman https://alliance.seas.upenn.edu/~cis520/wiki/images/cell_phone_face_detection.jpg http://maxcdn.liewcf.com/blog/wp-content/uploads/face-detection-camera-1.jpg Face detection http://glaucoma-eye-drops.com/eye.jpg – + - ≥0 http://glaucoma-eye-drops.com/eye.jpg – + - ≥0 Slightly better than random. http://glaucoma-eye-drops.com/eye.jpg http://graphics.stanford.edu/courses/cs148-11-fall/lectures/compression.pdf Combine weak classifiers to make a strong one http://www.codeproject.com/KB/audio-video/haar_detection/features1.png T (r; c) = X i·r;j·c I(i; j) ? +1 -1 -1 +1 +1 -1 -1 +1 +1 -1 -1 +1 Input: Weak classifiers hi (~x) 2 f¡1; 1g Output: StrongÃclassifier ! in form C(~x) = µ X i ®i hi (~x) Input: Weak classifiers hi (~x) 2 f¡1; 1g Output: StrongÃclassifier ! in form C(~x) = µ X i ®i hi (~x) All sub-windows Classifier 1 Not a face 50% Classifier 2 Not a face … Classifier 38 Not a face 2% Face 4916 positive examples, 9544 negative examples http://www.cs.ubc.ca/~lowe/425/slides/13-ViolaJones.pdf http://opencv.willowgarage.com/wiki/OpenCVLogo Context-Based Search for 3D Models Fisher and Hanrahan 2010 Characterizing Structural Relationships in Scenes Using Graph Kernels Fisher, Savva, and Hanrahan 2011 Individual Joint Joint Shape Segmentation with Linear Programming Huang, Koltun, and Guibas 2011 A Probabilistic Model of Component-Based Shape Synthesis Kalogerakis et al. 2012 http://www.cs.tau.ac.il/~wolf/OR/img/street_annotated.jpg CS 148, Summer 2012 Introduction to Computer Graphics and Imaging Justin Solomon