R-MVSNet Pytorch

Pytorch implementation of R-MVSNet.

This repo uses pytorch ported weights of original author's tensorflow implementation.

Installation

pip install git+git://github.com/leejaeyong7/rmvsnet-pytorch.git

Usage

from rmvsnet import RMVSNet

'''

Args: 
  images: Nx3xHxW tensor. H, W should be multiple of 16
  intrinsics: Nx3x3 tensor
  extrinsics: Nx4x4 tensor
  depth_start: float
  depth_interval: float
  depth_num: float

  depth ranges are computed by: depth_start + range(depth_num) * depth_interval

Return:
  probs: tensor of shape (H/4)x(W/4)
  depths: tensor of shape (H/4)x(W/4)
'''

model = RMVSNet()

# optional: put model into gpu
model.to(torch.device('cuda:0'))

depths, probs = model(images, intrinsics, extrinsics, depth_start, depth_interval, depth_num)

Reference:

About

This is a custom port of Original MVSNet using Tensorflow in Pytorch. We use same weight that the authors provided (GRU + DTU).

@article{yao2019recurrent,
  title={Recurrent MVSNet for High-resolution Multi-view Stereo Depth Inference},
  author={Yao, Yao and Luo, Zixin and Li, Shiwei and Shen, Tianwei and Fang, Tian and Quan, Long},
  journal={Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}