DS0707 - Interactions des mondes physiques, de l'humain et du monde numérique

Representation and manipulation of highly detailed shapes – RichShape

RichShape: Representation and manipulation of highly detailed shapes

During the last decade, we observed tremendous improvements in the quality of digitally generated images. This is mainly due to the availability of an ever increasing amount of shape details, and sophisticated appearance effects including complex lighting environments. This quest for extremely detailed shapes is further motivated by the generalization of very high resolution displays (e.g., 4k films, retina displays, etc.) which require extremely detailed images.

The challenge of high frequency meso geometries and micro-structures

In the context of computer graphics, this project aims to enhance the realism and visual richness of digital images. More precisely, our goal is to design novel representations for the efficient rendering of highly detailed 3D objects. First, the focus will be put on the representation of meso­scale geometry to find a lightweight and more controllable alternative to the classical bitmap­ based displacement, normal, albedo or bidirectional reflectance textures. Then, we plan to generalize our approach to reproduce very high­frequency effects produced by some classes of micro­scale geometry (e.g., brushed metals). Such effects are challenging to reproduce as they cannot be explicitly represented but smoothed out by classical material representations. Finally, we also plan to investigate procedural geometric details generation to go beyond what can be actually stored using explicit geometries.

In order to reach our goals, we aim to develop lightweight representations storing the minimal amount of information only. In order to reach real-time performances, we also aim to exploit as best as possible the capacity of nowadays and future GPUs. This implies numerous constraints on the design of our representations. In order to take into consideration the repetitive nature of some meso/micro geometric details, we aim to study semi-procedural approaches based on multi-layer distributions of primitives that will be combined on the fly at rendering time.

During this project, our primary result is a novel pipeline for the generation and efficient rendering of Level-Of-Details (LOD) compatible with hardware tessellation. To this end, we first showed how to overrule the limitations of the fixed HW tessellation pattern to enable non-uniform LOD inside patches. Then, we show how to generate LOD meshes compatible with these predefined HW tessellation pattern, and proposed a novel view-dependent LOD selection metric.

In parallel to this work, we also worked on the high-quality rendering of finely scratched materials such as metals, plastics, or finish woods. To this end, we first assumed a uniformly and regularly scratched surface to design a dedicated reflectance model, which is efficiently pre-computed through an optimized 2D ray- tracer taking into account all inter-reflections inside a scratch, including Fresnel effects. The outcome is compactly stored in a three-component 2D texture. Our approach provides users with controls over the profile and micro-BRDF of scratches, while updating our material model at interactive rates. This approach is simple, fast, and enables a large variety of finely detailed appearance with a very small amount of memory consumption.

Finally, as a requirement for the generation of high quality LOD compatible with HW tessellation, we also designed a novel solver of the 2D L2 optimal transport problem that is about 1 to 2 orders of magnitudes faster than state of the art solvers.

In the future it would be interesting to integrate our LOD generation and rendering pipeline to existing rendering engines. Typical applications would include video games, virtual reality for a more realistic immersion, or even online 3D model inspection if we combine our approche with compression scheme.

Optimal transport is a fundamental problem in numerous application domains and we thus expect that our novel solver will benefit way beyond the scope of this ANR project.

Main publications :

1. T. Lambert, P. Be´nard, G. Guennebaud. Multi-Resolution Meshes for Feature-Aware Hardware Tessellation. Computer Graphics Forum (Proc. of Eurographics 2016). 2016.

2. B. Raymond, G. Guennebaud, P. Barla. Multi-Scale Rendering of Scratched Materials using a Structured SV-BRDF Model. ACM Transactions on Graphics (Proc. of SIGGRAPH 2016). 2016.

In the context of computer graphics, this project aims to enhance the realism and visual richness of digital images. More precisely, our goal is to design novel representations for the creation, manipulation and efficient rendering of highly detailed 3D objects. First, the focus will be put on the representation of meso-scale geometry to find a lightweight and more controllable alternative to the classical bitmap-based displacement, normal, albedo or bidirectional reflectance textures. Then, we plan to generalize our approach to reproduce the very high-frequency effects produced by some classes of micro-scale geometry (e.g., brushed metals). Such effects are challenging to reproduce as they cannot be explicitly represented but smoothed out by classical material representations. Finally, we also plan to embed multi-scale aspects into our novel representation to enable alias-free rendering from distant viewpoints while avoiding the over-smoothing produced by current approaches.

Project coordination

gael GUENNEBAUD (Inria Bordeaux)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

Inria Inria Bordeaux

Help of the ANR 198,640 euros
Beginning and duration of the scientific project: September 2014 - 36 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter