一般的なオープンソースマルチビュー立体アルゴリズム実行スクリプトレコード
24373 ワード
1. MVE
プロジェクトのホームページhttps://www.gcc.tu-darmstadt.de/home/proj/mve/
Githubアドレスhttps://github.com/simonfuhrmann/mve
2. SMVS
プロジェクトのホームページhttps://www.gcc.tu-darmstadt.de/home/proj/smvs/smvs.en.jsp
Githubアドレスhttps://github.com/flanggut/smvs
3. openMVG+openMVS
マルチビュー立体ジオメトリベースライブラリopenMVGhttps://github.com/openMVG/openMVG
稠密再構築ライブラリopenMVShttps://github.com/cdcseacave/openMVS
4. COLMAP
プロジェクトのホームページhttps://demuc.de/colmap/
Githubアドレスhttps://github.com/colmap/colmap
colmap recon script(from tanks and temples)
上記のいくつかのスクリプトは私のgithubに置かれています.アドレスは次のとおりです.https://github.com/philleer/program_test/tree/mvs_script
Ubuntu 18.04 LTSは自らテストして有効で、補充を歓迎します
プロジェクトのホームページhttps://www.gcc.tu-darmstadt.de/home/proj/mve/
Githubアドレスhttps://github.com/simonfuhrmann/mve
#!/bin/bash
workspace_path=/root/test_result/mve_result
image_dir=${workspace_path}/${1}
scene_dir=${workspace_path}/${2}
mve=/root/misc_codes/mve/apps
maxpixel=20000000
intrinsic_fp="2759.48,0,0,0.4950,0.4916,0.9983" # fountain-p11
intrinsic_tp="1520.40,0,0,0.4724,0.5143,0.9964" # temple
intrinsic_eth_pipe="3430.27,0,0,0.5015,0.4969,1.0003" # eth3d pipes
intrinsic_dtu="2892.33,0,0,0.5145,0.5159,1.0032" # dtu dataset
intrinsic_tanks=""
intrinsic=${intrinsic_tanks}
# --init-intrinsics=${intrinsic} \
${mve}/makescene/makescene --original \
--images-only ${image_dir} \
--max-pixels=${maxpixel} \
${scene_dir} &&
# --fixed-intrinsics \
${mve}/sfmrecon/sfmrecon --max-pixels=${maxpixel} \
--verbose-ba ${scene_dir} &&
${mve}/dmrecon/dmrecon --neighbors=9 \
--scale=0 \
--max-pixels=${maxpixel} \
--local-neighbors=6 \
--keep-dz \
--progress=fancy ${scene_dir} &&
${mve}/scene2pset/scene2pset -F0 ${scene_dir} ${scene_dir}/pset-L0.ply &&
${mve}/fssrecon/fssrecon ${scene_dir}/pset-L0.ply ${scene_dir}/surface-L0.ply &&
${mve}/meshclean/meshclean --threshold=8.0 --delete-scale ${scene_dir}/surface-L0.ply ${scene_dir}/surface-clean.ply
2. SMVS
プロジェクトのホームページhttps://www.gcc.tu-darmstadt.de/home/proj/smvs/smvs.en.jsp
Githubアドレスhttps://github.com/flanggut/smvs
#!/bin/bash
workspace_path=/root/test_result/smvs_result
image_dir=${workspace_path}/${1}
scene_dir=${workspace_path}/${2}
mve=/root/misc_codes/mve/apps
smvs=/root/misc_codes/smvs/smvsrecon
maxpixel=2000000
intrinsic_fp="2759.48,0,0,0.4950,0.4916,0.9983" # fountain-p11
intrinsic_tp="1520.40,0,0,0.4724,0.5143,0.9964" # temple
intrinsic_eth_pipe="3430.27,0,0,0.5015,0.4969,1.0003" # eth3d pipes
intrinsic_dtu="2892.33,0,0,0.5145,0.5159,1.0032" # dtu dataset
intrinsic_tanks="2304.00,0,0,0.5,0.5,1.0000" # Manually set intrinsic
intrinsic=${intrinsic_fp}
${mve}/makescene/makescene --original \
--images-only ${image_dir} \
--max-pixels=${maxpixel} \
--init-intrinsics=${intrinsic} \
${scene_dir} &&
${mve}/sfmrecon/sfmrecon --max-pixels=${maxpixel} \
--fixed-intrinsics \
--verbose-ba ${scene_dir} &&
${smvs} ${scene_dir} &&
${mve}/fssrecon/fssrecon ${scene_dir}/pset-L0.ply ${scene_dir}/surface-L0.ply &&
${mve}/meshclean/meshclean --threshold=8.0 --delete-scale \
${scene_dir}/surface-L0.ply \
${scene_dir}/surface-clean.ply
3. openMVG+openMVS
マルチビュー立体ジオメトリベースライブラリopenMVGhttps://github.com/openMVG/openMVG
稠密再構築ライブラリopenMVShttps://github.com/cdcseacave/openMVS
#!/bin/bash
workspace_path=/root/test_result/openmvs_result/${1}
image_dir=${workspace_path}/images
recon_dir=${workspace_path}/reconstruct
match_dir=${recon_dir}/matches
openmvg=/root/misc_codes/openMVG/openmvg-bin/bin
openmvs=/root/misc_codes/openMVS/openmvs-build/bin
maxres=6400
minres=480
intrinsic_fp="2759.48;0;1520.69;0;2764.16;1006.81;0;0;1" # fountain-p11
intrinsic_tp="1520.40;0;302.32;0;1525.90;246.87;0;0;1" # temple
intrinsic_eth_pipe="3430.27;0;3119.2;0;3429.23;2057.75;0;0;1" # eth3d pipes
intrinsic_dtu="2892.33;0;823.21;0;2883.17;619.07;0;0;1" # for all dtu datasets
intrinsic_tanks="2304.00;0;960;0;2304.00;540;0;0;1" # Manually set intrinsic
intrinsic=${intrinsic_dtu}
mkdir ${recon_dir} && mkdir ${match_dir}
${openmvg}/openMVG_main_SfMInit_ImageListing -i ${image_dir} -o ${match_dir} \
--camera_model 1 \
--intrinsics ${intrinsic} \
--group_camera_model 1
${openmvg}/openMVG_main_ComputeFeatures -i ${match_dir}/sfm_data.json \
--outdir ${match_dir} \
--describerPreset HIGH
${openmvg}/openMVG_main_ComputeMatches -i ${match_dir}/sfm_data.json \
--out_dir ${match_dir} \
--nearest_matching_method ANNL2
${openmvg}/openMVG_main_IncrementalSfM -i ${match_dir}/sfm_data.json \
--matchdir ${match_dir} \
--outdir ${recon_dir} \
--camera_model 1 \
--refineIntrinsics NONE
# --refineIntrinsics "ADJUST_FOCAL_LENGTH|ADJUST_PRINCIPAL_POINT"
${openmvg}/openMVG_main_ComputeSfM_DataColor -i ${recon_dir}/sfm_data.bin \
-o ${recon_dir}/colorized.ply
${openmvg}/openMVG_main_ComputeStructureFromKnownPoses -i ${recon_dir}/sfm_data.bin \
--match_dir ${match_dir} \
--match_file ${match_dir}/matches.f.bin \
--output_file ${recon_dir}/robust.bin
${openmvg}/openMVG_main_ComputeSfM_DataColor -i ${recon_dir}/robust.bin \
-o ${recon_dir}/robust_colorized.ply
# outfile is the file name to save converted result
# outdir is the path to save undistorted images
${openmvg}/openMVG_main_openMVG2openMVS --sfmdata ${recon_dir}/sfm_data.bin \
--outfile ${recon_dir}/scene.mvs \
--outdir ${recon_dir}
${openmvs}/DensifyPointCloud --working-folder ${recon_dir} \
-i ${recon_dir}/scene.mvs \
--max-resolution=${maxres} \
--min-resolution=${minres} \
--number-views=6
# free-space-support is for textureless region
${openmvs}/ReconstructMesh --working-folder ${recon_dir} \
-i ${recon_dir}/scene_dense.mvs \
--free-space-support 1
${openmvs}/RefineMesh --working-folder ${recon_dir} \
-i ${recon_dir}/scene_dense_mesh.mvs \
--min-resolution ${minres} \
--max-views 9 \
--scales 5 \
--planar-vertex-ratio 5
${openmvs}/TextureMesh --working-folder ${recon_dir} \
-i ${recon_dir}/scene_dense_mesh.mvs \
--min-resolution ${minres} \
--cost-smoothness-ratio 0.3
4. COLMAP
プロジェクトのホームページhttps://demuc.de/colmap/
Githubアドレスhttps://github.com/colmap/colmap
#!/bin/bash
colmap=/root/misc_codes/colmap/colmap-bin/bin/colmap
workspace=/root/test_result/colmap_result/${1}
images=${workspace}/images
database_path=${workspace}/database.db
sparse_path=${workspace}/sparse
dense_path=${workspace}/dense
maxsize=2000
maxfeature=8192
intrinsic_fp="2759.48,2764.16,1520.69,1006.81" # fountain-p11
intrinsic_tp="1520.40,1525.90,302.32,246.87" # temple
intrinsic_dtu="2892.33,2883.17,823.21,619.07" # all dtu dataset
intrinsic_tanks=""
intrinsic=${intrinsic_dtu}
# --ImageReader.camera_params ${intrinsic} \
${colmap} feature_extractor \
--database_path ${database_path} \
--image_path ${images} \
--ImageReader.camera_model PINHOLE \
--ImageReader.camera_params ${intrinsic} \
--ImageReader.single_camera 1 \
--SiftExtraction.max_image_size ${maxsize} \
--SiftExtraction.max_num_features ${maxfeature}
${colmap} exhaustive_matcher --database_path ${database_path} \
--SiftMatching.guided_matching 0
mkdir ${sparse_path}
${colmap} mapper --database_path ${database_path} \
--image_path ${images} \
--output_path ${sparse_path} \
--Mapper.ba_refine_principal_point false
mkdir ${dense_path} &&
${colmap} image_undistorter --image_path ${images} \
--input_path ${sparse_path}/0 \
--output_path ${dense_path} \
--output_type COLMAP \
--max_image_size ${maxsize} &&
${colmap} patch_match_stereo --workspace_path ${dense_path} \
--workspace_format COLMAP \
--PatchMatchStereo.max_image_size ${maxsize} \
--PatchMatchStereo.window_radius 9 \
--PatchMatchStereo.geom_consistency 1 \
--PatchMatchStereo.filter_min_ncc 0.07 &&
${colmap} stereo_fusion --workspace_path ${dense_path} \
--input_type geometric \
--output_path ${dense_path}/fused.ply &&
${colmap} poisson_mesher --input_path ${dense_path}/fused.ply \
--output_path ${dense_path}/meshed-poisson.ply
${colmap} delaunay_mesher --input_path ${dense_path} \
--input_type dense \
--output_path ${dense_path}/meshed-delaunay.ply
colmap recon script(from tanks and temples)
#!/bin/bash
# ----------------------------------------------------------------------------
# - TanksAndTemples Website Toolbox -
# - http://www.tanksandtemples.org -
# ----------------------------------------------------------------------------
# The MIT License (MIT)
#
# Copyright (c) 2017
# Arno Knapitsch
# Jaesik Park
# Qian-Yi Zhou
# Vladlen Koltun
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
# THE SOFTWARE.
# ----------------------------------------------------------------------------
#
# This script generates a COLMAP reconstruction from a numbe rof input imagess
# Usage: sh get_colmap_reconstruction.sh
colmap_folder=$1/
iname=$2/
outf=$3/
DATABASE=${outf}sample_reconstruction.db
PROJECT_PATH=${outf}
mkdir -p ${PROJECT_PATH}
mkdir -p ${PROJECT_PATH}/images
cp -n ${iname}*.jpg ${PROJECT_PATH}/images
${colmap_folder}/colmap feature_extractor \
--database_path ${DATABASE} \
--image_path ${PROJECT_PATH}/images \
--ImageReader.camera_model RADIAL \
--ImageReader.single_camera 1 \
--SiftExtraction.use_gpu 1
${colmap_folder}/colmap exhaustive_matcher \
--database_path ${DATABASE} \
--SiftMatching.use_gpu 1
mkdir ${PROJECT_PATH}/sparse
${colmap_folder}/colmap mapper \
--database_path ${DATABASE} \
--image_path ${PROJECT_PATH}/images \
--output_path ${PROJECT_PATH}/sparse
mkdir ${PROJECT_PATH}/dense
${colmap_folder}/colmap image_undistorter \
--image_path ${PROJECT_PATH}/images \
--input_path ${PROJECT_PATH}/sparse/0/ \
--output_path ${PROJECT_PATH}/dense \
--output_type COLMAP --max_image_size 1500
${colmap_folder}/colmap patch_match_stereo \
--workspace_path $PROJECT_PATH/dense \
--workspace_format COLMAP \
--PatchMatchStereo.geom_consistency true
${colmap_folder}/colmap stereo_fusion \
--workspace_path $PROJECT_PATH/dense \
--workspace_format COLMAP \
--input_type geometric \
--output_path $PROJECT_PATH/dense/fused.ply
上記のいくつかのスクリプトは私のgithubに置かれています.アドレスは次のとおりです.https://github.com/philleer/program_test/tree/mvs_script
Ubuntu 18.04 LTSは自らテストして有効で、補充を歓迎します