Scalable Inside-Out Image-Based Rendering

摘要

Our aim is to give users real-time free-viewpoint rendering of realrnindoor scenes, captured with off-the-shelf equipment such as a highqualityrncolor camera and a commodity depth sensor. Image-basedrnRendering (IBR) can provide the realistic imagery required at realtimernspeed. For indoor scenes however, two challenges are especiallyrnprominent. First, the reconstructed 3D geometry must be compact,rnbut faithful enough to respect occlusion relationships when viewedrnup close. Second, man-made materials call for view-dependent texturing,rnbut using too many input photographs reduces performance.rnWe customize a typical RGB-D 3D surface reconstruction pipeline tornproduce a coarse global 3D surface, and local, per-view geometry forrneach input image. Our tiled IBR preserves quality by economizingrnon the expected contributions that entire groups of input pixels makernto a final image. The two components are designed to work together,rngiving real-time performance, while hardly sacrificing quality. Testingrnon a variety of challenging scenes shows that our inside-out IBRrnscales favorably with the number of input images.