search

TruongKhang / cds-mvsnet

[ICLR2022] Curvature-guided dynamic scale networks for Multi-view Stereo
117 stars 7 forks source link

aggregated cost V #8

Closed zhang-snowy closed 2 years ago

zhang-snowy commented 2 years ago

Why aggregated cost V（Cost V olume Formulation）, as in the paper? Does it mean anything?

TruongKhang commented 2 years ago

Hi @zhang-snowy, please take a look at our paper. Also, we recommend you a list of earlier works that use this concept, including VisMVSNet, PVSNet, or PatchMatchNet. I can simply explain as follows. First, we compute a two-view matching cost volume for each pair (reference image, source image). Then all two-view cost volumes from all pairs of reference-source image are integrated into a single volume V by using the weighted mean operation. Estimating the cost volume V is a common step in the multi-view stereo problem

zhang-snowy commented 2 years ago

Thank u : ) ！！！