Geometry Image-based Shadow Volume Algorithm for Subdivision Surfaces Min Tang, Jin-Xiang Dong College of Computer Science, Zhejiang University {tang_m,
Contents Related Works Related Works Main Contributions Main Contributions Algorithm Pipeline Algorithm Pipeline Evaluation Evaluation Silhouette Detection Silhouette Detection Implementation and Results Implementation and Results Demo Demo Conclusion & Future Works Conclusion & Future Works
Related Works Subdivision Surface Evaluation on GPU GPU based shadow volume algorithm Geometry Image The work here is a extension of our previous work at CGI06: Real-time Shadow Volume Algorithm for Subdivision Surface Based Models by Min Tang, et al.
Main Contributions Improved efficiency: By using two geometry images to pack all the patches’ information, the efficiencies of silhouette detection and shadow volume generation are greatly improved (2 render-to-texture ops per model under dynamic lighting). Improved accuracy: An enhanced silhouette detection method is designed to improve accuracy and ensure robustness
Algorithm Pipeline
Evaluation (a) Control meshes (b) Rendering using the two geometry images (c) The geometry image storing normal vectors (d) The geometry image storing limit positions (a) Control meshes (b) Rendering using the two geometry images
Silhouette Detection Directed visibility interpolation for silhouette extraction (+ for visible, - for invisible)
Silhouette Detection (a) Silhouette detection using normal vector of adjacent facets in [17] (b) Silhouette detection using the directed visibility interpolation method
Silhouette Detection Cracks caused by numeric instability [17] Smooth linked silhouettes
Silhouette Detection Cracks caused by numeric instability [17] Smooth linked silhouettes
Silhouette Detection Shadow leaks caused by the cracks [17] Our result with the same light source
Silhouette Rendring (a) a silhouette image(b) silhouettes
Shadow Volume Generation
Comparison with [17] Render-to-TextureFPS Our method 1~75 [17]Num_of_Patches*3~15
Implementation and Results (a) Shadows and silhouettes of the mode in Figure 1 (288 patches), ~75fps (b) shadows and silhouettes of a face-shaped model (288 patches), ~75fps (c) shadow and silhouettes of a complex model (480 patches), ~37fps
Demo&Source Code w/subShadow.htmhttp:// w/subShadow.htm.
Conclusion & future works Extension to geometry images Speed is limited by the RTVA efficiency
Thanks ! Q&A