Figure 1: Processing result for the objects described as Catmull-Clark subdivision surfaces in a scene from our demo featuring on-chip tessellation of subdivision surfaces (highlighted by the outlines of their control meshes). Animated curtain (on the left) is adaptively tessellated.
AbstractSubdivision surfaces possess many appealing properties applicable to interactive computer graphics. However, the necessity to access a variable-sized neighborhood in a control mesh makes it difficult to efficiently accelerate tessellation calculations in graphics hardware. The paper addresses this problem in two ways. First, it proposes a simple and inexpensive calculation scheme for the tessellation of Catmull-Clark subdivision surfaces which can be implemented in a geometry shader. It operates on the shader's vertex input only and does not require external texture memory access or multi-pass processing for tessellation. Second, the paper presents an extension to the post-transform and lighting (T'n'L) vertex cache operation that efficiently accelerates the processing of variable-size primitives serving as input for the geometry shader. We demonstrate on-chip tessellation of Catmull-Clark subdivision surfaces on an embedded hardware implementation. The described calculation scheme will be implementable on desktop hardware if limitations on the maximal input primitive size for the geometry shader are relaxed.
The one-shot approach, DeepMark, for fast clothing detection as a modification of a multi-target network, Center-Net [1], is proposed in the paper. The state-of-the-art accuracy of 0.723 mAP for bounding box detection task and 0.532 mAP for landmark detection task on the DeepFash-ion2 Challenge dataset [2] were achieved. The proposed architecture can be used effectively on the low-power devices.
The function representation (FRep) allows for construction of quite complex shapes such as isosurfaces of real-valued functions composed using the functionally defined primitives and operations, Calculating such functions in the complex cases can be very time-consuming. Extraction and visualization of isosurfaces for them can hardly be imagined interactive. In this paper we present a method of an interactive navigation through a "sculpture garden" containing non-intersecting FRep objects defined in the terms of specialized high-level language HyperFun. Before the actual isosurface extraction and visualization occurs, the objects are voxelized on a regular 3 0 grid with a possibility of a further adaptive voxelization. Polygonization employs a hierarchical representation of the voxelized data and a view-dependent isosurface reconstruction at the different levels of detail. To speedup the extraction process, an isosurface is constructed only in the visible part of the dataset with its updates performed incrementally as observer moves. Due to low preprocessing costs required for isosurface mesh construction, it is possible to visualize timedependent objects, If hardware is capable to calculate updates in real time.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.