An Adaptive Regular Structure for Web
6.4 Quad-based Multiresolution Structure
6.5.4 Device compatibility considerations
There are a variety of GPU chipsets from different vendors present in current mobile devices. From those, the most extended ones include the PowerVR SGX 5xx GPUs present in Apple devices and many Android phones, and the Qualcomm Adreno family integrated in Android HTC devices and many of the latest high-end Android devices. The hardware present in those two GPU families fully support OpenGL ES 2.0 and, in their latest versions, OpenGL ES 3.0. Unfortunately, while on Android devices the drivers typically expose VTF capabilities,the drivers from iOS 4.x and on does not. Rendering this feature unavailable in iPhone/iPad/iPod devices. The latest T6xx version of the ARM Mali GPU, integrated in the Nexus 10, also includes support for VTF. A number of tablet devices, in addition, use chipsets with discrete or integrated GPUs originally designed for netbooks, which fully support VTFs. An example are chips from the AMD Radeon HD series used in a number of Windows-based tablets.
6.6 Discussion
We have introduced a remote rendering approach, which builds a compact image-based multiresolution representation suitable for efficient distribution and rendering of highly detailed 3D models on low-powered platforms and, in particular, scripted environments such as Web browsers.
In the future, we want to study the impact of more aggressive lossy compression on the error introduced in geometry reconstruction.
Advantages. The pipeline is fully automatic and targets densely tessellated mod-els, such as those created by 3D scanning or modeling systems such as ZBrush.
Our approach bridges the gap that currently exists from general-purpose meshes to rendering oriented structures based on real-time tessellation with normal/bump maps, which are typical of modern gaming platform but currently require con-siderable human effort to create. The simplicity of a regularly re-meshed repre-sentation has many benefits. In particular it reduces random memory accesses and eliminates the indirection and storage of per triangle vertex indices and per vertex texture coordinates. The resulting representation is compact, can be built on top of existing image representations, and is very well suited to streaming.
Chapter 6.Adaptive Quad Patches: An Adaptive Regular Structure for Web Distribution and
Adaptive Rendering of 3D Models
82
Due to the negligible run-time CPU overhead, real-time performance is achieved both on conventional GPU platforms using OpenGL, and on the emerging web-based environments web-based on WebGL. Promising applications of the technology range, thus, from the automatic creation of rapidly renderable objects for local and online games to the set-up of browsable 3D models repositories in the web.
Limitations. The proposed method is not general purpose, but targets meshes defining closed objects with large components (i.e., typical solid objects without fine topological details). As for other compressed streamable formats, we do not strive to exactly replicate the original geometry and color, but only to visually approximate them in a faithful way. As a result, and similarly to compressed video/image formats, our representation is lossy, and thus not applicable in situations where precise measures of the original geometry are required (e.g., CAD systems).
Scalability. This technique performs vertex displacement of a base surface using the information contained in the image-encoded representation, thus Vertex-Texture-Fetch is a required feature. Although nowadays most desktop and mobile
GPUs already support texture access in the Vertex Shader (thanks to unified shader architectures), there are still some cases where it is not supported (i.e., old NVIDIA Tegra GPUs).
6.7 Bibliographical Notes
The major part of the content of this chapter was based on paper [Gobb 12], where we presented the Adaptive Quad Patches approach. Introduction and discussion are based on paper [Bals 13b], where we discussed both the approach presented in this chapter and the Compact Adaptive TetraPuzzles which is presented in Chapter5.
C 7 . . . .
ExploreMaps: Effi-cient Construction of Panoramic View Graphs of Complex 3D Environ-ments
In previous chapters, we have presented some approaches for interactive exploration of highly detailed 3D models on common 3D platforms. When dealing with scenes with complex lighting, on the other hand, real-time constraints impose hard limits on the achievable quality. The approach presented in this chapter addresses these limitations by introducing a simple graph repre-sentation named ExploreMaps, where nodes are nicely placed point of views, called probes, and arcs are smooth paths between neighboring probes. The basic idea is to ensure visual quality in all kinds of environments by presenting precomputed imagery rather than real-time-renderings. This chapter focuses on the problem of how to appropriately place these probes in arbitrary 3D environments and how to compute nice-looking path, while Chapter10will focus on the navigation interface approach than can be realized on top of the graph-based representation.
7.1 Introduction
R
research efforts have produced systems capable of rendering mod-erately complex environments [Yoon 08]. Besides of that, real-time constraints limit the quality that can be achieved to simple shading and/or baked illumination, specially on low performance platforms such as mobile and web environments.In current 3D repositories, such as Blend Swap, 3D Café or Archive3D, 3D models available for download are mostly presented through a few user-selected static images, or simple orbiting of simplified versions of the original 3D models.
83
Chapter 7.ExploreMaps: Efficient Construction of Panoramic View Graphs of Complex 3D
Environments
84
for interactive applications, exploration is typically limited to simplified models.
This is particularly true on low-powered mobile devices or script-based Internet browsers.
In this work, we introduce an approach aimed at automatically providing a richer experience in presenting 3D models on low-powered platforms and web environments. The method builds on a novel efficient technique for transform-ing a generic renderable 3D scene into a simple graph representation, dubbed ExploreMaps, where nodes are nicely placed panoramic views, called probes, and arcs are smooth panoramic video paths connecting neighboring probes.
Our GPU-accelerated unattended construction pipeline distributes probes so as to guarantee complete coverage of a generic scene, before clustering them using perceptual criteria, determining preferential viewing orientations, finding smooth good looking connection paths. Probe images and path videos are then computed with off-line photo-realistic renderers, overcoming real-time rendering limitations. At run-time, the graph is exploited both for generating visual scene indexes and movie previews, and for supporting interactive exploration through a low-DOF assisted navigation interface (see Chapter 10 for details). Due to negligible CPU/GPU usage, real-time performance is achieved on emerging WebGL environments even on low-powered mobile devices (see Fig.7.1).
Figure 7.1: ExploreMaps pipeline.We automatically transform a generic renderable model (left) into a simple graph representation named ExploreMaps (center), where nodes are nicely placed point of views that cover the visible model surface and arcs are smooth paths between neighboring probes. The representation is exploited for providing visual indexes for the 3D scene and for supporting, even on low-powered mobile devices, interactive photo-realistic exploration based on precomputed imagery (right).