Download presentation
Presentation is loading. Please wait.
Published byDaisy Wilkerson Modified over 9 years ago
1
Bayesian Networks for Sketch Understanding Christine Alvarado MIT Student Oxygen Workshop 12 September 2003
2
Sketching in Design Mechanical Engineering Software
3
Computer Design Tools
4
A Challenge In Sketch Understanding Noisy Input There is no one threshold for shapes or constraints Interpretation depends on context
5
Naïve Approach Why not just try all possibilities?
6
Naïve Approach Why not just try all possibilities? Arrow?
7
Naïve Approach Why not just try all possibilities? Arrow?
8
Naïve Approach Why not just try all possibilities? Arrow?
9
Naïve Approach Why not just try all possibilities? Arrow?
10
Naïve Approach Why not just try all possibilities? Arrow?
11
Naïve Approach Why not just try all possibilities? Must consider interpretations n = number of strokes/segments S = set of shapes k i = subcomponents in shape S i
12
Naïve Approach Why not just try all possibilities? Must consider interpretations n = number of strokes/segments S = set of shapes k i = subcomponents in shape S i And this only considers shapes independently
13
Previous Approaches Use Rigid Segmentation Single stroke shapes Palm Pilot Graffiti Long et. al. [1999] Explicit Segmentation Quickset: Cohen et. al. [2001] Pause between strokes
14
Recognition Using Partial Interpretations Recognition: Build partial interpretations (PIs) as the user draws based on easily recognizable low-level shapes Prune unlikely PIs and use likely PIs to find misrecognized low-level shapes Evaluating PIs Graphical Models: Missing data = unobserved nodes Interpretation influenced by top-down and bottom-up information
15
BN fragments [similar to PRMs, Getoor et. al. 1999] (Define Arrow (Components (Line shaft) (Line head1) (Line head2)) (Constraints (connects shaft.p1 head1.p1) (connects shaft.p1 head2.p1) (= head1.length head2.length) (< head1.length shaft.length) (< (angle head1 shaft) 90) (< (angle shaft head2) 90) (> (angle head1 shaft) 0) (> (angle shaft head2) 0))) L1: L2: L3: C1: C2: C3: C4: C5: C6: C7: C8: Arrow L1L2L3C1C2C3C8 … Instantiated and linked together as recognition proceeds [Hammond and Davis, 2003]
16
Specifying Conditional Probability Tables P(Child | Parent) specified in description: P(L1 | Arrow) = 1 Probability of “optional” components/constraints lower Parents combine with noisy-OR (XOR) Priors specified for top-level shapes/patterns Arrow L1L2L3C1C2C3C8 … QuadrilateralArrow L1
17
Primitive shapes/Constraints Observation node added when primitive linked to stroke P(Obs|Prim) determined through data collection L1 Obs
18
Example Sq. error (Stroke a) Sq. error (Stroke b) Line(l1) Connects l1 l2 Arrow Line(l2)Line(l3) Quad Line Remaining Arrow Constraints Force(F) Force-pushes-body Body(B) Touches F B 0.99 0.95 0.97 0.99 0.5 0.59 Observation
19
Example Sq. error (Stroke a) Sq. error (Stroke b) Sq. error (Stroke c) Line(l1) Connects l1 l2 Arrow Line(l2)Line(l3) Quad Line Remaining Arrow Constraints Force(F) Force-pushes-body Body(B) Touches F B 0.99 0.95 0.97 0.99 0.5 0.59 Observation
20
Example Sq. error (Stroke a) Sq. error (Stroke b) Sq. error (Stroke c) Line(l1) Connects l1 l2 Arrow Line(l2) Line(l3) Quad Line Remaining Arrow Constraints Force(F) Force-pushes-body Body(B) Touches F B 1 1 1 0.97 1 0.47 0.61 0.95 Observation
21
Example Sq. error (Stroke a) Sq. error (Stroke b) Sq. error (Stroke c) Line(l1) Connects l1 l2 Arrow Line(l2) Line(l3) Quad Line Remaining Arrow Constraints Force(F) Force-pushes-body Body(B) Touches F B Observation 1 1 1 1 1 0.47 0.95 1 Sq. error (Stroke d) Ellipse 0.99 0.97 0.99 Observation
22
Uh oh… what about speed? Networks get very large (and ugly) Solutions Prune network Assert values (even if not confirmed) Approximate inference Incremental changes to inference structures
23
Current/Future Work Expand domain/include other domains Gather sketches from users
24
Conclusion Graphical models evaluation Partial Interpretations Context-guided search More drawing freedom + More robust recognition = More natural interfaces (i.e. The goal of OXYGEN)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.