ExampleBased Fractured Appearance L Glondu L Muguercia M
Example-Based Fractured Appearance L. Glondu, L. Muguercia, M. Marchal, C. Bosch, H. Rushmeier, G. Dumont and G. Drettakis
Introduction • Reproducing a specific fracture pattern with simulation can be a tedious task Photograph Simulation • Obtaining the same pattern is often not necessary • Similarity between two fracture patterns not well defined
Our goal • Estimate fracture simulation parameters from photographs (of cracks) • Match statistics rather than exact patterns – Fragment areas, edge lengths, junctions, … • Determine which statistics influence similarity – Use this as a metric during optimization • Adapt existing simulation approach
Contributions • User study to determine similarity between fracture patterns based on statistics • Optimization process using this metric • Extended RT fracture simulation approach • Interactive modeling interface
Related work (I) • Aging and weathering – Physically-based simulation [DORSEY et al. 1996, MÉRILLOU et al. 2008] – Data driven [WANG et al. 2006, GU et al. 2006] – Example-based simulation (stains) [BOSCH et al. 2011]
Related work (II) • Fracture simulation – – Continuum mechanics [O’BRIEN et al. 1999, MULLER et al. 2004] RT + implicit approach [GLONDU et al. 2012] Stress field [IBEN and O’BRIEN 2006] Procedural [MOULD 2005, GOBRON and CHIBA 2001, DESBENOIT et al. 2005]
Related work (III) • Statistical models for fracture – Validate simulation [VALETTE et al. 2008] – Matching of fragments [SHIN et al. 2010]
Outline 1. 2. 3. 4. 5. Processing of input patterns User study on statistical pattern similarity Fracture simulation approach Optimization of parameters Results
Processing input patterns 1) Extract cracks location 2) Process features: fragments, edges, junctions 3) Obtain statistics: areas, lengths, angles, … Photograph Cracks Features Statistics
User study (I) Which simulated pattern is the most similar to the reference pattern ? (2 -AFC) Reference pattern (from photo) Choice 1 (simulation) Choice 2 (simulation)
User study (II) • 5 reference images • 7 simulation conditions – – Matching fragment statistics (S 1) Matching crack statistics (S 2) Matching junction statistics (S 3) Combinations of S 1, S 2 and S 3 (S 1+S 2, S 2+S 3, …) • 20 participants (14 males, 6 females) • 212 comparisons (~20 min. )
Images for the user study
Demo
User study results S 1 S 2 S 3 S 1 + S 3 91 79. 5 50 27 22 S 2 9 S 3 20. 5 73 S 1 + S 3 50 78 27. 5 72. 5 • Fragment statistics (S 1) as predominant choice (except S 1 + S 3) • Statistics relevant for defining visual similarity
Fracture simulation • FEM based on modal analysis [GLONDU et al. 2012] – Fast + volumetric approach • Stress map evolving over time [IBEN and O’BRIEN 2006] – Initial stress computed from loading forces – Opening based on resistance threshold Rc (per-element) t=0 (no stress) t > 0 resulting fracture
Crack propagation • Implicit fracture surface [GLONDU et al. 2012] • We incorporate stress relaxation around cracks – Smoothing kernel based on distance (+ radius) – Combine with tensor to model preferred orientations Omni-directional relaxation Directional relaxation
Relaxation results Omni-directional relaxation (no preferred orientation) Directional relaxation (preferred orientations) • Selection based on input pattern statistics (mean junction angle)
Parameters to optimize Parameter symbol Description Loading force magnitude Relative to Body mass Stress increase rate Resistance to fracture variance Resistance to fracture Stress relaxation rate Resistance to fracture Stress relaxation radius • Input material properties: E, v, ρ, Rcm • Others: age, path noise
Optimization
Error metric • Fragment statistics Crack statistics Junction statistics EMD : Earth Mover’s Distance (wf, wc, wj) : weighting coefficients = (3, 1, 1) based on user study
Optimization approach • Approximate Bayesian Computation [BEAUMONT et al. 02] – Takes into account randomness of Rc and force position • Run N = 20 k simulations (~15 min. ) • Take the parameter set with lowest error
Results
Complex scenes
Fracture evolution Increasing age parameter
Volumetric approach True 3 -D fracture surfaces
Performance • Simulation – Timings: 16 ms - 264 ms – Model size: 7 K elements (tile) to 54 K (road) • Interactive editor: 30 -100 fps • Optimization: 15 min. (60 ms/iteration) Intel Core 2 Extreme, 2. 3 GHz, 4 GB RAM + n. Vidia Quadro FX 3700 M
Conclusion • Fracture similarity metric based on statistics and user study – 2 D statistics better than 1 D • Optimization method for fitting parameters based on this metric • Efficient simulation allowing interactive application of fractured appearance • Main limitation: no internal information – Thickness, texture, …
Thank you! Acknowledgments • TIN 2010 -20590 -C 02 -02 grant • NSF-1064412 grant • Donations: Autodesk, Adobe, n. Vidia
- Slides: 29