Submitted by Fangjinhua Wang.

Submission data

Full namePatchmatchNet
DescriptionWe present PatchmatchNet, a novel and learnable cascade formulation of Patchmatch for high-resolution multi-view stereo. With high computation speed and low memory requirement, PatchmatchNet can process higher resolution imagery and is more suited to run on resource limited devices than competitors that employ 3D cost volume regularization. For the first time we introduce an iterative multi-scale Patchmatch in an end-to-end trainable architecture and improve the Patchmatch core algorithm with a novel and learned adaptive propagation and evaluation scheme for each iteration. Extensive experiments show a very competitive performance and generalization for our method on DTU, Tanks & Temples and ETH3D, but at a significantly higher efficiency than all existing top-performing models: at least two and a half times faster than state-of-the-art methods with twice less memory usage.
Parametersimage size: 2688*1792
Publication titlePatchmatchNet: Learned Multi-View Patchmatch Stereo
Publication authorsFangjinhua Wang, Silvano Galliani, Christoph Vogel, Pablo Speciale, Marc Pollefeys
Publication URLhttps://arxiv.org/abs/2012.01411
Programming language(s)python
HardwareIntel(R) Core(TM) i7-9700K CPU @ 3.60GHz, GeForce RTX 2080, RAM: 32GB
Source code or download URLhttps://github.com/FangjinhuaWang/PatchmatchNet
Submission creation date8 Aug, 2020
Last edited18 Mar, 2021

High-res multi-view results



Infoallhigh-res
multi-view
indooroutdoorbotani.boulde.bridgedoorexhibi.lectur.living.loungeobserv.old co.statueterrac.
permissive73.1271.3378.4983.1860.8579.6378.5764.1371.7379.8151.2085.9757.4076.3688.66

Low-res many-view results



Infoalllow-res
many-view
indooroutdoorlakesidesand boxstorage roomstorage room 2tunnel
No results yet.

Low-res two-view results



Infoalldeliv. area 1ldeliv. area 1sdeliv. area 2ldeliv. area 2sdeliv. area 3ldeliv. area 3select. 1lelect. 1select. 2lelect. 2select. 3lelect. 3sfacade 1sforest 1sforest 2splayg. 1lplayg. 1splayg. 2lplayg. 2splayg. 3lplayg. 3sterra. 1sterra. 2sterra. 1lterra. 1sterra. 2lterra. 2s
No results yet.

SLAM results



allcables 1cables 2cables 3camera shake 1camera shake 2camera shake 3ceiling 1ceiling 2desk 3desk changing 1einstein 1einstein 2einstein darkeinstein flashlighteinstein global light changes 1einstein global light changes 2einstein global light changes 3kidnap 1kidnap darklarge loop 1mannequin 1mannequin 3mannequin 4mannequin 5mannequin 7mannequin face 1mannequin face 2mannequin face 3mannequin headmotion 1planar 2planar 3plant 1plant 2plant 3plant 4plant 5plant darkplant scene 1plant scene 2plant scene 3reflective 1repetitivesfm benchsfm gardensfm house loopsfm lab room 1sfm lab room 2sofa 1sofa 2sofa 3sofa 4sofa dark 1sofa dark 2sofa dark 3sofa shaketable 3table 4table 7vicon light 1vicon light 2
MethodInfo
No results yet.