ghixu/acmh

A simple yet effective PatchMatch MVS method, which is the base model of ACMM and ACMP.

ACMH

[News] The code for ACMM is released!!!
[News] The code for ACMP is released!!!
[News] The code for ACMMP is released!!!

About

ACMH is a simple yet effective PatchMatch MVS method, which is the base model of ACMM, ACMP and ACMMP. If you find this project useful for your research, please cite:

@article{Xu2019ACMM,  
  title={Multi-Scale Geometric Consistency Guided Multi-View Stereo}, 
  author={Xu, Qingshan and Tao, Wenbing}, 
  journal={Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}
@article{Xu2020ACMP,  
  title={Planar Prior Assisted PatchMatch Multi-View Stereo}, 
  author={Xu, Qingshan and Tao, Wenbing}, 
  journal={AAAI Conference on Artificial Intelligence (AAAI)},
  year={2020}
}

Dependencies

The code has been tested on Ubuntu 14.04 with GTX Titan X.

• Cuda >= 6.0
• OpenCV >= 2.4
• cmake

Usage

• Complie ACMH

cmake .  
make

• Test

Use script colmap2mvsnet_acm.py to convert COLMAP SfM result to ACMH input   
Run ./ACMH $data_folder to get reconstruction results

Acknowledgemets

This code largely benefits from the following repositories: Gipuma and COLMAP. Thanks to their authors for opening source of their excellent works.

acmh's Issues

I have some problems

Memory out of bounds in ComputeMultiViewInitialCostandSelectedViews?

Hello! In ComputeMultiViewInitialCostandSelectedViews，the initial size of the array cost_vector and cost_vector_copy is 32. If the params.num_images is greater than 32, memory out of bounds will occur. Right?

invalid bilateral space sigma parameter

In the code settings

the bilateral space weighting is almost truncated flat part of Gaussian distribution.
only 15% from the peak to valley, shown by the top image.

Even set the same weight in the kernel, no visual difference from the result.

The above weight distribution is more space logic?
The idea is 3 * sigma cover the 99% Gaussian.

min = 0.8681234453945849
max = 1.0
max/min = 1.151909910168909

import cv2
import numpy as np

# ksize  = 31
ksize  = 11
center = ksize // 2
# sigma  = ksize / 3 / 2 / 1.6627
sigma  = 5
A = np.zeros((ksize, ksize), dtype=float)

for y in range(ksize):
    for x in range(ksize):
        distX = x - center
        distY = y - center
        dist  = np.sqrt(distX*distX + distY*distY)
        v     = np.exp(-dist / (2*sigma*sigma))
        A[y, x] = v

minV = np.min(A[:])
maxV = np.max(A[:])
print(minV, maxV, maxV/minV)

A = cv2.resize(A, (256, 256))

cv2.imshow('img', A)
cv2.waitKey()

跑出来最终的结果不是点云？

I run failed

an illegal memory access was encountered

Hello! Sometimes there will be such an error: "an illegal memory access was encountered" in ACMH.cu 1105. Need help! Thank you.