WIKI de Bertrand VANDEPORTAELE

sudo apt-get install libc6-dev-i386 libhdf5-serial-dev libgtest-dev libgsl0-dev python-dev cmake imagemagick libmagick++-dev gfortran minpack-dev liblapack-dev libatlas-dev libatlas-base-dev libboost-dev libc6-dev-i386 libgsl0-dev libgtk2.0-dev libglew1.6-dev libglew1.6 libdevil-dev libboost-all-dev libatlas-cpp-0.6-dev libatlas-dev imagemagick libatlas3gf-base libcminpack-dev libgfortran3 libmetis-edf-dev libparmetis-dev freeglut3-dev libgsl0-dev

ou sous ubuntu 15.04

sudo apt-get install libc6-dev-i386 libhdf5-serial-dev libgtest-dev libgsl0-dev python-dev cmake imagemagick libmagick++-dev gfortran minpack-dev liblapack-dev libatlas-dev libatlas-base-dev libboost-dev libc6-dev-i386 libgsl0-dev libgtk2.0-dev libglew1.6-dev glew-utils libdevil-dev libboost-all-dev libatlas-cpp-0.6-dev libatlas-dev imagemagick libatlas3gf-base libcminpack-dev libgfortran3 libmetis-edf-dev libparmetis-dev freeglut3-dev libgsl0-dev

ou sous ubuntu 16.04

sudo apt-get install libc6-dev-i386 libhdf5-serial-dev libgtest-dev libgsl0-dev python-dev cmake imagemagick libmagick++-dev gfortran minpack-dev liblapack-dev libatlas-dev libatlas-base-dev libboost-dev libc6-dev-i386 libgsl0-dev libgtk2.0-dev libglew1.6-dev glew-utils libdevil-dev libboost-all-dev libatlas-cpp-0.6-dev libatlas-dev imagemagick libatlas3-base libcminpack-dev libgfortran3 libmetis-edf-dev libparmetis-dev freeglut3-dev libgsl0-dev f2c

cd ~
mkdir vsfm
cd vsfm

#récupération de SiftGPU-V400.zip sur http://wwwx.cs.unc.edu/~ccwu/cgi-bin/siftgpu.cgi

wget http://wwwx.cs.unc.edu/~ccwu/cgi-bin/siftgpu.cgi 
mv siftgpu.cgi SiftGPU-V400.zip 
unzip SiftGPU-V400.zip
cd SiftGPU/
nano makefile
  ajouter apres le 1° endif:  siftgpu_enable_cuda = 0
make
mkdir -p ../vsfm/bin 
cp bin/libsiftgpu.so ../vsfm/bin
cd ..

#la version 1.0.4 n'existe plus pba_v1.0.4.zip (1021 KB)

wget http://grail.cs.washington.edu/projects/mcba/pba_v1.0.5.zip
unzip pba_v1.0.5.zip
cd pba
mv makefile makefile_gpu
cp makefile_no_gpu makefile
make
cd ..

#comme j'ai compilé la version sans gpu et que visualsfm compte utiliser la version avec, il faut que j'adapte le nom de la lib binaire…. à tester…

cd pba/bin
ln -s libpba_no_gpu.so libpba.so
cd ../../

wget http://www.di.ens.fr/pmvs/pmvs-2.tar.gz
tar xf pmvs-2.tar.gz  
cd pmvs-2/program/main/
cp mylapack.o mylapack.o.backup
make clean
cp mylapack.o.backup mylapack.o
make depend
make
cd ../../../

#depuis http://www.cs.utexas.edu/users/dml/Software/graclus.html

wget http://www.cs.utexas.edu/users/dml/Software/graclus1.2.tar.gz
tar xvf graclus1.2.tar.gz 
cd graclus1.2/
nano Makefile.in
 set “-DNUMBITS=64″ 
make
cd ..

wget http://ccwu.me/vsfm/download/VisualSFM_linux_64bit.zip
unzip VisualSFM_linux_64bit.zip 
cd vsfm 
make
bin/VisualSFM
cd ..

wget http://www.di.ens.fr/cmvs/cmvs-fix2.tar.gz
tar xf cmvs-fix2.tar.gz
cp pmvs-2/program/main/mylapack.o cmvs/program/main/
nano cmvs/program/base/cmvs/bundle.cc
   #ajouter au début
      #include <vector>
      #include <numeric>
nano cmvs/program/main/genOption.cc
   #ajouter au début
      #include <stdlib.h>
nano cmvs/program/main/Makefile
   #Your INCLUDE path (e.g., -I/usr/include)
   YOUR_INCLUDE_PATH =
   #Your metis directory (contains header files under graclus1.2/metisLib/)
   YOUR_INCLUDE_METIS_PATH = -I/home/bvandepo/vsfm/graclus1.2/metisLib
   #Your LDLIBRARY path (e.g., -L/usr/lib)
   YOUR_LDLIB_PATH = -L/home/bvandepo/vsfm/graclus1.2

cd ~/vsfm/cmvs/program/main
make
cp cmvs ~/vsfm/vsfm/bin
cp pmvs2 ~/vsfm/vsfm/bin
cp genOption ~/vsfm/vsfm/bin
cd ~/vsfm

archivage des fichiers d'installation:

mkdir archive
mv cmvs-fix2.tar.gz graclus1.2.tar.gz pba_v1.0.5.zip pmvs-2.tar.gz SiftGPU-V400.zip VisualSFM_linux_64bit.zip archive

ajouter dans ~/.bashrc:

nano ~/.bashrc
  export PATH=$PATH:/home/bvandepo/vsfm/vsfm/bin
  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/bvandepo/vsfm/vsfm/bin:/home/bvandepo/vsfm/SiftGPU/bin/:/home/bvandepo/vsfm/pba/bin/
source ~/.bashrc

echo $LD_LIBRARY_PATH

pour lancer:

VisualSFM &

en cas de problème de librairie:

ldd ./vsfm/bin/VisualSFM

documentation http://ccwu.me/vsfm/doc.html#

installation de meshlab pour visualiser les modèles .ply

sudo apt-get install meshlab

Utilisation

en interface graphique

File → open +multi images→charger les pgm

sfm →pairwise matching→ compute missing matches

sfm → reconstruct sparse

selectionner les caméras qui posent probleme

bouton main grise ?→delete selected camera

sfm→pairwise matching→show spanning forest et matrix

bouton triangle rouge et bleu: compute 3D reconstruction

bouton cmvs⇒ choisir dossier /tmp/img sur ramdisk et donner un nom au fichier

view→dense 3D points

si il y a un problème lors du bundle à cause du choix des 2 images de références, relancer le bundle après avoir fait:

Sfm->Pairwise Matching->Update Pairwise F-Matrix

sauvegarde du résultat en nvm ou ply

Sfm->Extra Function->Save Current Model

au cas ou il y aurait plusieurs modèles générés

Sfm->More options->Merge sparse model   puis relancer BA

Sfm->More options->Find more points    puis relancer BA

pour afficher les correspondances entre les images:

View->feature match


Resume 3D reconstruction peut enlever des mauvais matches

Imposer des paramètres intrisèques fixés:

Sfm->More function->Set fixed calibration

format attendu

fx cx fy cy r
2605.73990 1364.50522 2612.01336 983.24601 0

en ligne de commande

~/cartography/sequences/enac_ulysse/vol2$ VisualSFM -h

+sort:    keep the input image order in the output NVM;
+fixcam:  keep existing cameras unmodified.

Exemple de séquence traitée

https://bvdp.inetdoc.net/files/vsfm/

Génération des images et listes

script.sh

#!/bin/bash 
IMG_DIR="select/"
mkdir -p $IMG_DIR
LIST_FILE=$IMG_DIR"list.txt"
rm $LIST_FILE
rm $IMG_DIR/image_*.pgm
rm $IMG_DIR/image_*.png
rm $IMG_DIR/image_*.jpg
rm $IMG_DIR/image_*.tif
rm $IMG_DIR/image_*.sift
rm $IMG_DIR/image_*.mat
rm $IMG_DIR/*.nvm
#copie des images avec des noms consécutifs
j=0
#paramètres de seq   image de départ, incrément entre chaque image, image de fin
#for i in  $(seq -w 1 40 3769);   do
#for i in  $(seq -w 100 3 400);   do
for i in  $(seq -w 1 30 3769);   do
#formate sur 4 digits
if=`printf "%04.f\n" $i`
jf=`printf "%04.f\n" $j`
echo $if" -> "$jf
cp image_$if.pgm $IMG_DIR/image_$jf.pgm
#convert  image_$i.pgm select/image_$k.tif
j=`expr $j + 1` 
done
#génération de la liste des paires d'images
for k in  $(seq -w 1 120);   do
for inc in  $(seq  1 5);   do
l=`expr $k + $inc` 
#formatage sur 4 digits
kf=`printf "%04.f\n" $k`
lf=`printf "%04.f\n" $l`
echo image_$kf.pgm image_$lf.pgm  >>$LIST_FILE
done
done

script2.sh

#!/usr/bin/env /bin/bash
 
USER_DATA=select
OUTPUT_DATA=select/results
OUTPUT_NVM=${OUTPUT_DATA}.nvm
 
if [ "$1" = "clean" ]; then
  rm -rf $OUTPUT_NVM ${USER_DATA}/*.mat ${USER_DATA}/*.sift
  echo "Done cleaning. Leaving"
  exit 0
fi
 
HOMEREP=/home/bvandepo/
export PATH=$HOMEREP/vsfm/cmvs/program/main:$HOMEREP/vsfm/vsfm/bin:$PATH
export LD_LIBRARY_PATH=$HOMEREP/vsfm/pba/bin/:$HOMEREP/vsfm/vsfm/bin:$HOMEREP/vsfm/SiftGPU/bin/:$LD_LIBRARY_PATH
 
#VisualSFM sfm+pmvs+sort $USER_DATA/list.txt $OUTPUT_NVM
 
VisualSFM sfm+pmvs+sort list.txt result.nvm
 
VisualSFM sfm+sort list.txt result.nvm
 
 
VisualSFM sfm+sort   result.nvm

Autre

options.txt

  VisualSFM Version = 0.5.26
 
  ------------------------------------------------
  VisualSFM sfm[options] input output.nvm [user_data_path]
  options = [match_option][sfm_option][misc_option][mvs_option]
  match_option = [+pairs/+import/+subset/+nomatch/]
    +pairs:   match image pairs from [user_data_path];
              use @range,step as path for sequence mode;
    +import:  load feature matches from [user_data_path]; 
    +subset:  enable preemptive feature matching when
              #images >= param_preemptive_switch;
    +nomatch: reconstruction without feature matching;
    default:  compute missing pairwise matches. 
  sfm_option = [+resume/+add/+skipsfm/+loadnvm/]
    +resume:  load NVM file, try add new images from 
              [input].txt, and grow the existing models;
    +add:     load NVM file, and find more points and
              projections for the existing models;
    +loadnvm: load NVM file and skip feature matching;
    +skipsfm: skip sparse and dense reconstruction;
    default:  run regular sparse reconstruction.
  misc_option = [+k=fx,cx,fy,cy/+shared][+sort][+gcp][+fixcam][+merge]
    +k:       fixed calibration, e.g. +k=1024,800,1024,600;
    +shared:  enforce shared calibration in the end;
    +sort:    keep the input image order in the output NVM;
    +gcp:     load [input].gcp and transform the 3D model.
    +fixcam:  keep existing cameras unmodified.
    +merge:   try model merging after sparse reconstruction
  mvs_option = [+pmvs/+cmvs/+cmp/]
    +cmvs:    undistort images, run CMVS/genOption, skip PMVS;
    +pmvs:    undistort images, run CMVS/genOption/PMVS;
    +cmp:     undistort images, write p-matrices for CMP-MVS;
    default: skip the entire dense reconstruction.
  The <input> can be the following: 
    a .txt file containing relative/absolute image paths,
    a folder path to look for .jpg files, or
    an existing NVM file.
 
  ------------------------------------------------
  VisualSFM listen[+log] port 
    Start the GUI and listen to a socket port for UI commands,
    which simulate menu clicks. Each command should in the form 
    of "id[c][s] [argument]\n". The [id] is an integer that 
    corresponds to a menu item, for which the complete list 
    can be found in the sample ui.ini file. The optional flag 
    [c][s] corresponds CTRL and SHIFT press; The [argument] is 
    for the input string like filepath.
    Log messages will be send to client if +log is specified
   puis relancer BA
  ------------------------------------------------
  VisualSFM sfm[export_op   puis relancer BAtions] input output.txt
    +exportp: export putative feature matches
    +exportf: export F-matrix inlier matches
 
  ------------------------------------------------
  VisualSFM siftgpu[+ascii] image_list_file
    Detect features for a list of images

example_utilisation.txt

 [1] 21712
bvandepo@garossos:~$ CPU single-precisoin solver; 2 cores.
Reading bundle...40 cameras -- 22512 points in bundle file
***********
40 cameras -- 22512 points
Reading images: ****************************************
Set widths/heights...done	0 secs
done	0 secs
slimNeighborsSetLinks...done	0 secs
mergeSFM...***********resetPoints...done
Rep counts: 22512 -> 509  	1 secs
setScoreThresholds...done	0 secs
sRemoveImages... **********
Kept: 3 4 5 6 7 8 9 10 12 13 16 17 18 19 23 24 25 26 28 29 30 31 32 33 34 35 36 37 38 39 
 
Removed: 0 1 2 11 14 15 20 21 22 27 
sRemoveImages: 40 -> 30	0 secs
slimNeighborsSetLinks...done	0 secs
Cluster sizes: 
30 
Adding images: 
0 
Image nums: 40 -> 30 -> 30
Divide: 
done	0 secs
27.8667 images in vis on the average
--------------------------------------------------
--- Summary of specified options ---
# of timages: 30 (enumeration)
# of oimages: 0 (enumeration)
level: 1  csize: 2
threshold: 0  wsize: 7
minImageNum: 3  CPU: 2
useVisData: 1  sequence: -1
--------------------------------------------------
Reading images: ******************************
3 Harris running ...4 Harris running ...1876 harris done
DoG running...2010 harris done
DoG running...2699 dog done
5 Harris running ...2638 dog done
6 Harris running ...1946 harris done
DoG running...1978 harris done
DoG running...2654 dog done
7 Harris running ...2095 harris done
DoG running...2653 dog done
8 Harris running ...2693 dog done
9 Harris running ...2000 harris done
DoG running...2050 harris done
DoG running...2598 dog done
10 Harris running ...2637 dog done
12 Harris running ...1990 harris done
DoG running...1796 harris done
DoG running...2692 dog done
13 Harris running ...2644 dog done
16 Harris running ...1862 harris done
DoG running...1948 harris done
DoG running...2728 dog done
17 Harris running ...2567 dog done
18 Harris running ...1910 harris done
DoG running...2061 harris done
DoG running...2680 dog done
19 Harris running ...2688 dog done
23 Harris running ...1970 harris done
DoG running...2015 harris done
DoG running...2699 dog done
24 Harris running ...2600 dog done
25 Harris running ...1992 harris done
DoG running...1822 harris done
DoG running...2551 dog done
26 Harris running ...2662 dog done
28 Harris running ...1942 harris done
DoG running...2079 harris done
DoG running...2702 dog done
29 Harris running ...2010 harris done
DoG running...2610 dog done
30 Harris running ...2567 dog done
31 Harris running ...1939 harris done
DoG running...1926 harris done
DoG running...2519 dog done
32 Harris running ...1964 harris done
DoG running...2604 dog done
33 Harris running ...2607 dog done
34 Harris running ...2046 harris done
DoG running...1984 harris done
DoG running...2633 dog done
35 Harris running ...2505 dog done
36 Harris running ...1966 harris done
DoG running...1974 harris done
DoG running...2634 dog done
37 Harris running ...1908 harris done
DoG running...2630 dog done
38 Harris running ...2604 dog done
39 Harris running ...1911 harris done
DoG running...1925 harris done
DoG running...2604 dog done
2555 dog done
done
adding seeds 
(4,4269)(28,4300)(11,3664)(15,4133)(16,1936)(24,4317)(17,1843)(26,3458)(29,4113)(9,1592)(22,1900)(19,351)(3,2927)(2,424)(23,332)(25,310)(18,0)(13,276)(8,291)(0,1252)(21,1434)(7,531)(12,13)(14,19)(20,5)(5,2004)(6,2)(10,2)(1,9)(27,5)done
---- Initial: 0 secs ----
Total pass fail0 fail1 refinepatch: 145847 88328 48965 8554 96882
Total pass fail0 fail1 refinepatch: 100 60.5621 33.5729 5.86505 66.4271
Expanding patches...
---- EXPANSION: 2829 secs ----
Total pass fail0 fail1 refinepatch: 924156 847591 58111 18454 866045
Total pass fail0 fail1 refinepatch: 100 91.7151 6.28801 1.99685 93.712
FilterOutside
mainbody: 
Gain (ave/var): 1.44734 2.87679
893293 -> 611767 (68.4845%)	0 secs
Filter Exact: ******************************
611767 -> 292769 (47.8563%)	0 secs
FilterNeighbor:	292769 -> 234085 (79.9555%)	0 secs
FilterGroups:	23
234085 -> 194522 (83.0989%)	0 secs
STATUS: 16614 0 946311 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 
Expanding patches...
---- EXPANSION: 2659 secs ----
Total pass fail0 fail1 refinepatch: 1074636 589061 108676 376899 965960
Total pass fail0 fail1 refinepatch: 100 54.8149 10.1128 35.0722 89.8872
FilterOutside
mainbody: 
Gain (ave/var): 2.28291 2.07301
783583 -> 713287 (91.0289%)	0 secs
Filter Exact: ******************************
713287 -> 351883 (49.3326%)	0 secs
FilterNeighbor:	351883 -> 320880 (91.1894%)	0 secs
FilterGroups:	32
320880 -> 276521 (86.1758%)	0 secs
STATUS: 38079 0 1890801 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 
Expanding patches...
---- EXPANSION: 2338 secs ----
Total pass fail0 fail1 refinepatch: 1045429 553496 105713 386220 939716
Total pass fail0 fail1 refinepatch: 100 52.9444 10.1119 36.9437 89.8881
FilterOutside
mainbody: 
Gain (ave/var): 2.28821 2.04883
830017 -> 754364 (90.8854%)	0 secs
Filter Exact: ******************************
754364 -> 368196 (48.8088%)	0 secs
FilterNeighbor:	368196 -> 338478 (91.9288%)	0 secs
FilterGroups:	33
338478 -> 292789 (86.5016%)	0 secs
STATUS: 58848 0 2809745 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 0 0 0 0 0 
0 0 0 0 0 
---- Total: 0 secs ----
 
 
 
 
 
Running Yasutaka Furukawa's CMVS tool...
cmvs /home/bvandepo/sample/mvs.nvm.cmvs/00/ 50 2
genOption /home/bvandepo/sample/mvs.nvm.cmvs/00/ 1 2 0,700000 7 3 2
1 cluster is generated for pmvs reconstruction.
pmvs2 /home/bvandepo/sample/mvs.nvm.cmvs/00/ option-0000
This may take a little more time, waiting...
8501 seconds were used by PMVS
Loading option-0000.ply, 292789 vertices ...
Loading patches and estimating point sizes..
 
#############################
You can manually remove bad MVS points:
1. Switch to dense MVS points mode;
2. Hit F1 key to enter the selection mode;
3. Select points by dragging a rectangle;
4. Hit DELETE to delete the selected points.
 
#############################
Save to mvs.nvm ... done
Save /home/bvandepo/sample/mvs.0.ply ...done
----------------------------------------------------------------
Run dense reconstruction, finished
Totally 142,467 minutes used

tuto en francais: http://combiencaporte.blogspot.fr/2012/07/la-photogrammetrie-visualsfm-et-meshlab.html

Meshlab

Filter→ Sampling / Poisson-disk Sampling.

La case Base mesh subsampling doit être cochée (car pour l'instant les éléments du maillage n'existent pas).
Number of samples: 10000
  Création d'un petit nuage de point contenant environ 10k points, en appliquant la méthode de rééchantillonnage vue ci-dessus. On applique ensuite sur ce petit échantillon

Filter→ Point Set / Compute Normals for Point Set,

avec la case Flip normals w.r.t. viewpoint cochée8, et en entrant les coordonnées d'un point fictif éloigné du maillage (par exemple 0,0,-1000).

Filters→ Remeshing, simplification and reconstruction → Surface Reconstruction : Poisson,

avec Octree Depth égal à 6. 
  Transférer ensuite les normales de ce maillage vers le nuages de points complets, avec

Filters→Sampling / Vertex Attribute Transfer

en cochant uniquement Transfer normal, et en choisissant votre premier maillage (poisson mesh?????) pour Source mesh et votre nuage de points complet pour Target mesh. Il est possible de remplacer ces deux premières étapes en appliquant directement le filtre Compute Normals for Point Set au nuage de points complet, si ce dernier est suffisamment petit. Si le nuage est trop grand, l'application du filtre peut-être très longue ou faire planter MeshLab.

Filters→Remeshing, simplification and reconstruction / Surface Reconstruction : Poisson,

en choisissant Octree Depth entre 6 et 12 selon la précision du maillage désiré, et Solver Divide entre la valeur de Octree Depth et cette valeur moins 2.
 Attention le temps de calcul augmente très rapidement avec Octree Depth, et il vaut mieux commencer avec Octree Depth égal à 6 et l'augmenter de 1 en 1. L'effet de la valeur de Octree Depth est illustrée sur la figure 4. Sur cet exemple on voit le raffinement du maillage au fur et à mesure que le paramètre Octree Depth augmente9. A la fin de l'étape (3), on obtient donc le maillage recherché, mais il ne possède pour l'instant pas de couleur.

Pour transférer les couleurs contenues dans le nuage de point sur votre nouveau maillage, appliquez Sampling / Vertex Attribute transfer, en cochant Transfer color, et en choisissant votre nuage de points pour Source mesh et votre nouveau maillage pour Target mesh.

Filter→ Point Set / Compute Normals for Point Set,

avec la case Flip normals w.r.t. viewpoint cochée8, et en entrant les coordonnées d'un point fictif éloigné du maillage (par exemple 0,0,-10000).

Problème, la moitié des normales est inversée….