ShadowDraw Real-Time User Guidance for Freehand Drawing

25
ShadowDraw Real-Time User Guidance for Freehand Drawing Larry Zitnick, Michael Cohen Microsoft Research Yong Jae Lee U. of Texas at Austin

description

ShadowDraw Real-Time User Guidance for Freehand Drawing. Larry Zitnick, Michael Cohen Microsoft Research. Yong Jae Lee U. o f Texas at Austin. Drawing. Subject 1. Subject 2. We need help. Tracing. Difficult to find photos Limits creativity. Drawing a face. Learn a set of rules - PowerPoint PPT Presentation

Transcript of ShadowDraw Real-Time User Guidance for Freehand Drawing

ShadowDraw Real-Time User Guidance for Freehand Drawing

ShadowDraw Real-Time User Guidance for Freehand DrawingLarry Zitnick, Michael CohenMicrosoft ResearchYong Jae LeeU. of Texas at AustinDrawing

Subject 1Subject 2We need help..Tracing

Difficult to find photosLimits creativity

Drawing a faceLearn a set of rulesRequires significant training

4

Our Idea: ShadowDrawShadowDrawCreating shadows

=ShadowShadow gives gist of many images simultaneously

+Collection of imagesCreating shadowsWhat if the only input is the users drawing?

ShadowDrawingRequires partial matching and dynamic updatesSpeedReal-time updating of shadows is critical

0.1 seconds feels right0.5 seconds useful2 seconds useless

OverviewUser DrawingFor each sub-windowHistogram(ID x dx x dy)Top 100Fine AlignmentSpatial ScoringShadow

VerifyQuery time (online)Database Image

EdgesFor each sub-windowSketch, Img_ID, x, y.Sketch, Img_ID, x, ySketch, Img_ID, x, ySketch, Img_ID, x, yMin Hash

Database (offline)Database

DatabaseEfficient (sub-linear) retrievalStores image index and offsetInverted file structure:

[Zitnick, ECCV 2010]

Edge descriptor:PositionOrientationLinear lengthDatabase Image

EdgesFor each sub-windowSketch, Img_ID, x, y.Sketch, Img_ID, x, ySketch, Img_ID, x, ySketch, Img_ID, x, yMin Hash

Database (offline)MatchingSecond stage: Refine alignment and compute blending weightsFirst stage: Use inverted file structure to find candidate set User DrawingFor each sub-windowHistogram(ID x dx x dy)Top 100

Fine AlignmentSpatial ScoringShadow

VerifyQuery time (online)Fine alignmentHough transform (3D)Break into three 1D problems:YXXYS (scale)SUser DrawingFor each sub-windowHistogram(ID x dx x dy)Top 100

Fine AlignmentSpatial ScoringShadow

Verify

Blending weights

=Shadows are the composite of many images**+Pen StrokesTop matchesWeights

Blending weights=**+

Shadows are the composite of many images

Top matchesWeightsPen Strokes

Rendering

+Pen StrokesPen PositionShadow

*+=OutputHigher contrast near pen positionMain focus is users drawingRelated workInteractive drawing interfacesTeddy [Igarashi et al., 1999]Fluid Sketches [Arvo and Novins, 2000]3D drawing system [Igarashi and Hughes, 2000]iCanDraw [Dixon et al., 2010]

Drawing studiesWhere do people draw lines? [Cole et al., 2008]

User studies30,000 images, 20 categories16 drawers (8 men, 8 women), 8 evaluators 5 objects (shoe, face, bicycle, butterfly, rabbit)

User studiesGood drawersWith ShadowDrawWithout ShadowDraw

User studiesBad drawersWith ShadowDrawWithout ShadowDraw

User studiesAverage drawersWith ShadowDraw

Without ShadowDrawUser studiesAverage drawers

With ShadowDrawWithout ShadowDrawUser studies

Significant improvement for Average groupUser studies

Improvement for most object categoriesRabbit is control variableUser studies

After training all users improved:Poor skillGood skillSubjects personal style is maintained!Future workMatching against drawings and not photographsTemporal strokesColor, shading, etc.Drawing priors

Not future work:Changing peoples strokesLive demo