Supertetras: A Superpixel Analog for Tetrahedral Mesh Oversegmentation

摘要: Over the past decade, computer vision algorithms have transitioned from relying on direct, pixel-based representation of images to use superpixels, small regions whose boundaries agree with image contours. This intermediate improves tractability understanding because it reduces number primitives be taken under consideration several million a few hundred. Despite improvements yielded in area segmentation, concept an oversegmentation as has not been adopted volumetric mesh processing. We take first step this direction, adapting fast and efficient superpixel algorithm tetrahedral case, present results which demonstrate quality output oversegmentation, illustrate its semantic segmentation application.