Abstract

We present efficient implementations of two primitives for datamapping and distribution on the massively multithreaded architec-ture of the GPUs in this paper. Thesplitprimitive distributes el-ements of a list according to their category. Split is an importantoperation for data mapping and is used to build data structures,distribute work load, etc., in a massively parallel environment.Thegather/scatterprimitive performs fast, distributed data move-ment. Efficient data movement is critical to high performance onthe GPUs as suboptimal memory accesses can pay heavy penalties.The split we implement is a generalization of the binary split [Blel-loch 1990] and is implemented using the shared memory and theatomic operations available on them. The split performance scaleslogarithmically with the number of categories, linearly with the listlength, and linearly with the number of cores on the GPU. Thismakes it useful for applications that deal with large data sets. Wealso present a variant of split that partitions the indexes of records.This facilitates the use of the GPU as a coprocessor for split or sort,with the actual data movement handled separately. We can computethe split indexes for a list of 32 million records in 180 millisecondsfor a 32-bit key and in 800 ms for a 96-bit key. The instantaneouslocality of memory references play a critical role in data movementon the current GPU memory architectures. For scatter and gatherinvolving large records, we use collective data movement in whichmultiple threads cooperate on individual records to improve the in-stantaneous locality. The split, gather, and their combinations findmany applications and expect our primitives will be used by fu-ture GPU programmers. We show sorting of 16 million 128-byterecords in 379 milliseconds with 4-byte keys and in 556 ms with8-byte keys

Abstract

The Graphics Processing Units (GPUs) provide highcomputation power at a low cost and is an important computeaccelerator with a massively multithreaded architecture. Inthis paper, we present fast implementations of common graphoperations like breadth-first search, st-connectivity, single-sourceshortest path, all-pairs shortest path, minimum spanning tree,and maximum flow for undirected graphs on the GPU using theCUDA programming model. Our implementations exhibit highperformance, especially on large graphs. We use two data-parallelprogramming methodologies for these algorithms. One is aniterative, mask-based approach that processes valid data elementslike vertices and edges using independent threads for each.The other is a divide-and-conquer approach that reduces theproblem into smaller problems that are handled later using thesame approach. Parallel algorithms for such problems have beenreported in the literature before, especially on supercomputers.The massively multithreaded model of the GPU makes it possibleto adopt the data-parallel approach even to irregular algorithmslike graph algorithms, usingO(V)orO(E)simultaneous threads.The algorithms and the underlying techniques presented in thispaper are likely to be applicable to many irregular algorithms.We show the impact of our implementations on random, scale-free, and real-life graphs of up to millions of vertices on high-end and low-end GPUs. The availability and spread of GPUs todesktops and laptops make them ideal candidates to accelerategraph operations over the CPU-only implementations. Practicalimplementations of basic operations go a long way in realizingtheir potential.

Abstract

A video from a moving camera produces different number of observations of different scene areas. We can construct an attention map of the scene by bringing the frames to a common reference and counting the number of frames that observed each scene point. Different representations can be constructed from this. The base of the attention map gives the scene mosaic. Super-resolved images of parts of the scene can be obtained using a subset of observations or video frames. We can combine mosaicing with superresolution by using all observations, but the magnification factor will vary across the scene based on the attention received. The height of the attention map indicates the amount of super-resolution for that scene point. We modify the traditional super-resolution framework to generate a varying resolution image for panning cameras in this paper. The varying resolution image uses all useful data available in a video. We introduce the concept of attention-based super-resolution and give the modified framework for it. We also show its applicability on a few indoor and outdoor videos.

Abstract

Video completion algorithms have concentrated on obtaining visually consistent solutions to fill-in the missing portions, without any emphasis on the physical correctness of the video. Resulting solutions thus use texture or image structure based cues and are limited in the situations they can handle. In this paper we take a model based signal processing approach to video completion [1]. Completion of the video is then defined as satisfying the given model by detecting and removing the error (selected parts of the video to be replaced). Given a probabilistic model, video completion then becomes an unsupervised learning algorithm with the input video giving a “noisy” version. Dense completion is the automatic inferencing of the “noise-less” or “true” video from the input. This approach finds a solution that satisfies visual coherence and is applicable to a wide variety of scenarios. We demonstrate the efficacy of our approach and its wide applicability using two scenarios.

Abstract

Many events are captured using multiple cameras today. Frames of each video stream have to be synchronized and aligned to a common time axis before processing them. Synchronization of the video streams necessarily needs a hardware based solution that is applied while capturing. The alignment problem between the frames of multiple videos can be posed as a search using traditional measures for image similarity. Multiview relations and constraints developed in Computer Vision recently can provide more elegant solutions to this problem. In this paper, we provide two solutions for the video frame alignment problem using two view and three view constraints. We present solutions to this problem for the case when the videos are taken using affine cameras and for general projective cameras. Excellent experimental results are achieved by our algorithms.

Abstract

We present a non-photo realistic, real-time painterly ren-dering technique for terrains. The painterly appearanceand the impression of terrains is created by effectively ren-dering several brush strokes. The strokes have fixed loca-tions on the surfaces of the terrain during animation to en-able frame to frame coherency. The strokes are rendered asalpha blended sprites in two-dimensions and are orientedalong the slope of terrain analogous to the way artists painton canvas. By exploiting the regular nature of terrain data,we create pre-decided rendering depth orders for primitivesfor any camera orientation.With this, we avoid the neces-sity of sorting the primitives of sprites required for alphablending. We use DirectX10/SM4.0 based shaders to renderstrokes to improve performance. Being distributed on ter-rain, strokes get cluttered when they are closely located onscreen. We follow a level of detail scheme that maintains auniform stroke density in screen space. Various styles canbe achieved with different stroke variations. Phong shadingthe rendered output in real-time is possible for more variedstyles. We achieve painterly rendering in real-time with acombination of object space positioning and image spacerendering of strokes. We illustrate our method with imagesand performance results.

Abstract

The primitives of point-based representations are in-dependent but are rendered using surfels, which approxi-mate the immediate neighborhood of each point linearly.A large number of surfels are needed to convey the exactshape. Higher-order approximations of the local neigh-borhood have the potential to represent the shape usingfewer primitives,simultaneously achieving higher renderingspeeds. In this paper, we proposealgebraic splatsas abasic primitive of representation for point based models.An algebraic splat based representation can be computedusing a moving least squares procedure. We specificallystudy low order polynomial splats in this paper. Quadraticand cubic splats provide good quality and high renderingspeed using far fewer primitives on a wide range of mod-els. They can also be rendered fast using ray tracing onmodern GPUs. We also present an algorithm to constructa representation of a model with a user-specified number ofprimitives. Our method to generates a hole-free represen-tation parametrized by a smoothing radius. The hole-freerepresentation reduces the number of primitives needed bya factor 20 to 30 on most models and by a factor of over100 on dense models like David with little or no drop invisual quality. We also present a two-pass GPU algorithmthat ray-traces the algebraic splats and blends them usinga Gaussian weighting scheme for smooth appearance. Weare able to render models like David at upwards of 200 fpson a commodity GPU using algebraic splats.

Abstract

The GPUs pack high computation power and a restricted architecture into easily available hardware today. They are now used as computation co-processors and come with programming models that treat them as standard paral-lel architectures. We explore the problem of real time raycasting of large deformable models (over a million trian-gles) on large displays (a million pixels) on an off-the-shelf GPU in this paper. Ray casting is an inherently parallel and highly compute intensive operation. We build a GPU-efficient three-dimensional data structure for this purpose and a corresponding algorithm that uses it for fast ray cast-ing. We also present fast methods to build the data struc-ture on the SIMD GPUs, including a fast multi-split opera-tion. We achieve real-time ray-casting of a million triangle model onto a million pixels on current Nvidia GPUs using the CUDA model. Results are presented on the data struc-ture building and ray casting on a number of models. The ideas presented here are likely to extend to later models and architectures of the GPU as well as to other multi core ar-chitectures.

Abstract

Terrains are challenging geometric objects for real-timerendering and interactive manipulation. State-of-the-artterrain rendering systems use custom, multiresolution, rep-resentations like geometry clipmaps for fast rendering onthe GPU. In this paper, we present a system that exploitsthe power and flexibility of the modern GPUs to store, ren-der, and manipulate terrains with minimal CPU involve-ment. The central idea is to use a regular-grid represen-tation and fixed size blocks/tiles that change in resolution.The potentially visible portion of the terrain is cached atthe highest necessary resolution and is rendered from theGPU. The CPU sends a light geometry template which isexpanded by the Geometry Shader to the triangles, usingthe heights stored in the GPU Cache. The CPU performsa coarse culling of the tiles with the GPU performing fineculling. The GPU cache is updated continuously as theviewpoint changes. Our system enables the terrain to bemodified procedurally or edited interactively on the GPUwith no CPU involvement. The terrain can also interactwith a large number of external objects in real-time entirelywithin the GPU. We achieve a consistent rendering rate of100 frames per second with terrain modification and inter-actions as well as a triangle rate of upto 350 million persecond on an Nvidia 8800 GTX GPU for large terrains, witha CPU load below 10%.

Abstract

Terrains can be rendered efficiently with rectangular 2D grid of heights. Terrain on spheres, on the other hand, can be rendered using Hierarchical Triangular Mesh (HTM) but the representation does not fit directly with 2D grid of heights. We present a unified representation of HTM and clip mapping using Hexagonal Geome-try Clipmaps. This provides one to one correspondence of vertices and heights, low and constant memory usage, less storage space requirements, no pole singularity, fast vertex look-ups, and large range of view distances. Hexagonal clip maps fit equilateral trian-gles of HTM and achieve uniform triangle count on the screen. We keep the clipmaps on the GPU for fast access from shaders and usevertex buffer objects for fast triangle rendering.