We present a computational framework for the inference of dense descriptions from multiple view stereo with general camera placement. Thus far research on dense multiple view stereo has evolved along three axes: computation of scene approximations in the form of visual hulls; merging of depth maps derived from simple configurations, such as binocular or trinocular; and multiple view stereo with restricted camera placement. These approaches are either suboptimal, since they do not maximize the use of available information, or cannot be applied to general camera configurations. Our approach does not involve binocular processing other than the detection of tentative pixel correspondences. We require calibration information for all cameras and that there exist camera pairs which enable automatic pixel matching. The inference of scene surfaces is based on the premise that correct pixel correspondences, reconstructed in 3-D, form salient, coherent surfaces, while wrong correspondences form less coherent structures. The tensor voting framework is suitable for this task since it can process the very large datasets we generate with reasonable computational complexity. We show results on real images that present numerous challenges.