1. Lightfields, Lumigraphs, and Other Image-Based Methods

2. Image-Based Modeling and Rendering For many applications, re-rendering is goalFor many applications, re-rendering is goal Traditional vision / graphics pipelines:Traditional vision / graphics pipelines: Image-based pipeline:Image-based pipeline: WorldWorld NewImagesNewImages Light sources ReflectanceReflectance GeometryGeometry VisionGraphics WorldWorld New Images CapturedImagesCapturedImages

3. Plenoptic Function L(x,y,z, , ,t, )L(x,y,z, , ,t, ) Captures all light flow in a sceneCaptures all light flow in a scene Enough information to construct any image of the scene at any timeEnough information to construct any image of the scene at any time Practical approximations:Practical approximations: – Static scenes: ignore dependence on t – Represent color as RGB: eliminate – Represent color as RGB: eliminate – 7D  3  5D

4. Plenoptic Function – Special Cases Sample at one (x,y,z):Sample at one (x,y,z): – L( ,  ) is just an (omnidirectional) image Full 5D L(x,y,z, ,  ):Full 5D L(x,y,z, ,  ): – Omnidirectional image at each point in space – Enough information to reconstruct any view

5. Free Space Consider a region of space without occlusionConsider a region of space without occlusion Light travels in straight lines  some pixels in different images are the same ray of lightLight travels in straight lines  some pixels in different images are the same ray of light “Rebinning” pixels General case Free space

6. Light Field In unoccluded space, can reduce plenoptic function to 4DIn unoccluded space, can reduce plenoptic function to 4D 2D array of 2D images2D array of 2D images Still contains enough information to reconstruct new viewsStill contains enough information to reconstruct new views

7. Image-Based Modeling and Rendering Generate new views of a scene directly from existing viewsGenerate new views of a scene directly from existing views “Pure” IBR (such as lightfields): no geometric model of scene“Pure” IBR (such as lightfields): no geometric model of scene Other IBR techniques try to obtain higher quality with less storage by building a modelOther IBR techniques try to obtain higher quality with less storage by building a model

8. Lightfields Advantages:Advantages: – Simpler computation vs. traditional CG – Cost independent of scene complexity – Cost independent of material properties and other optical effects – Avoid hard vision problems Disadvantages:Disadvantages: – Static geometry – Fixed lighting – High storage cost

9. Using Lightfields Obtain 2D slices of 4D data setObtain 2D slices of 4D data set Arbitrary views: take other 2D slicesArbitrary views: take other 2D slices Challenges:Challenges: – Capture – Parameterization – Compression – Rendering

10. Capturing Lightfields Need a 2D set of (2D) imagesNeed a 2D set of (2D) images Choices:Choices: – Camera motion: human vs. computer – Constraints on camera motion – Coverage and sampling uniformity – Aliasing

11. Point / anglePoint / angle Two points on a sphereTwo points on a sphere Points on two planesPoints on two planes Original images and camera positionsOriginal images and camera positions Lightfield Parameterization

12. Compression Compress individual images (JPEG, etc.)Compress individual images (JPEG, etc.) Adapt video compression to 2D arraysAdapt video compression to 2D arrays Decomposition into basis functionsDecomposition into basis functions Vector quantizationVector quantization

13. Rendering How to select rays?How to select rays? InterpolationInterpolation Taking advantage of hardwareTaking advantage of hardware – Graphics hardware – Compression hardware

14. Implementations Lightfields, Levoy and Hanrahan ( SIGGRAPH 96 )Lightfields, Levoy and Hanrahan ( SIGGRAPH 96 ) Lumigraphs, Gortler et al. ( SIGGRAPH 96 )Lumigraphs, Gortler et al. ( SIGGRAPH 96 ) Unstructured lumigraphs, Buehler et al. ( SIGGRAPH 01 )Unstructured lumigraphs, Buehler et al. ( SIGGRAPH 01 )

15. Light Field Rendering Capture:Capture: – Computer-controlled camera rig – Move camera to grid of locations on a plane

16. Light Field Rendering Parameterization:Parameterization: – Two planes, evenly sampled: “light slab” – In general, planes in arbitrary orientations – In practice, one plane = camera locations Minimizes resamplingMinimizes resampling

17. Light Field Coverage

18. Multi-Slab Light Fields

19. Rendering For each desired ray:For each desired ray: – Compute intersection with (u,v) and (s,t) planes – Take closest ray Variants: interpolationVariants: interpolation – Bilinear in (u,v) only – Bilinear in (s,t) only – Quadrilinear in (u,v,s,t)

20. Light Field Compression Based on vector quantizationBased on vector quantization Preprocessing: build a representative codebook of 4D tilesPreprocessing: build a representative codebook of 4D tiles Each tile in lightfield represented by indexEach tile in lightfield represented by index Example: 2x2x2x2 tiles, 16 bit index = 24:1 compressionExample: 2x2x2x2 tiles, 16 bit index = 24:1 compression

21. The Lumigraph Capture: move camera by handCapture: move camera by hand Camera intrinsics assumed calibratedCamera intrinsics assumed calibrated Camera pose recovered from markersCamera pose recovered from markers

22. Lumigraph Postprocessing Obtain rough geometric modelObtain rough geometric model – Chroma keying (blue screen) to extract silhouettes – Octree-based space carving Resample images to two-plane parameterizationResample images to two-plane parameterization

23. Lumigraph Rendering Use rough depth information to improve rendering qualityUse rough depth information to improve rendering quality

24. Lumigraph Rendering Use rough depth information to improve rendering qualityUse rough depth information to improve rendering quality

25. Lumigraph Rendering Without using geometry Using approximate geometry

26. Unstructured Lumigraph Rendering Further enhancement of lumigraphs: do not use two-plane parameterizationFurther enhancement of lumigraphs: do not use two-plane parameterization Store original pictures: no resamplingStore original pictures: no resampling Hand-held camera, moved around an environmentHand-held camera, moved around an environment

27. Unstructured Lumigraph Rendering To reconstruct views, assign penalty to each original rayTo reconstruct views, assign penalty to each original ray – Distance to desired ray, using approximate geometry – Resolution – Feather near edges of image Construct “camera blending field”Construct “camera blending field” Render using texture mappingRender using texture mapping

28. Unstructured Lumigraph Rendering Blending fieldRendering

29. Other Lightfield Acquisition Devices Spherical motion of camera around an objectSpherical motion of camera around an object Samples space of directions uniformlySamples space of directions uniformly Second arm to move light source – measure reflectanceSecond arm to move light source – measure reflectance

30. Other Lightfield Acquisition Devices Acquire an entire light field at onceAcquire an entire light field at once Video ratesVideo rates Integrated MPEG2 compression for each cameraIntegrated MPEG2 compression for each camera (Bennett Wilburn, Michal Smulski, Mark Horowitz)