ฝัง
- เผยแพร่เมื่อ 15 มิ.ย. 2024
- An overview of different software tools which can be used to obtain a 3D model from a set of photographs. This procedure (or parts of it) can go under the names "photogrammetry", "Structure from motion", and "Multi-view stereo".
The goal of this presentation is just to illustrate a few of these programs: what they can do, and how they can be used to quickly get a result with minimal user effort; it's not meant to be a technical comparison of these tools' effectiveness.
The photos used for the reconstruction in this video are publicly available:
github.com/rperrot/Reconstruc...
More information about the Snap packaging format:
snapcraft.io
Links to the tools:
VisualSFM - ccwu.me/vsfm/index.html
MeshLab - www.meshlab.net/
Bundler - www.cs.cornell.edu/~snavely/bu...
CMVS - www.di.ens.fr/cmvs/
COLMAP - colmap.github.io/
MVE - www.gcc.tu-darmstadt.de/home/p...
MVS-Texturing - www.gcc.tu-darmstadt.de/home/p...
openMVG - www.gcc.tu-darmstadt.de/home/p...
OpenMVS - cdcseacave.github.io/openMVS/
Theia - www.theia-sfm.org/ - วิทยาศาสตร์และเทคโนโลยี
Brilliant. Thanks for taking the time to put all this together. A lot of steep learning curves I’m sure. I’m a VisualSFM user, and now I will have to try some of the other options you show that I’ve heard about but never took the time to investigate. Well done!
C'est la synthèse la plus claire que j'ai pu voir sur les outils de photogrammétrie. Merci.
This is a brilliant video !! I've not come across such an in depth and complete review of open source photogrammetry software before thank you !!
Excellent video! Thank you. I've bookmarked this and will have to watch it many times to fully understand all of the details here and try it for myself. Thanks again.
Okay, I've tried each of these methods on a set of photos I have and I don't get as good of results from any of these methods as I do with Regard3D. But I have to run Windows because I can't get it to compile correctly in Linux. Any change you could create a Regard3D snap?
Great video and EXTRA points for the snaps. Thanks for taking all this effort.
Great list and thanks for the source images! Man it's crazy that how many good option exist.
thank you. great video. please continue to post more materials!
Great video Mardy! Thanks and keep it up! :D
And here I was building all these programs manually (srsly, because I'm a software developer), which is a serious pain (dependency madness + cmake quirks). Unfortunately, I didn't know that all I needed to do was: "sudo snap install visualsfm-mardy" (any idea how to alter the text size in this context?)
As a result, I did learn a lot of internal details about many of them, and your pipeline schematic is quite helpful. Thanks!
01:39 Overview
02:51 VisualSFM
04:07 Meshlab
09:18 COLMAP
11:30 Bundler
13:03 CMVS
14:18 MVE
17:30 openMVG
19:48 openMVS
21:38 Theia
Fantastically helpful video, thank you
thanks so much for this video. really appreciate taking the time
It's the most informative videos I have seen in the internet. I have one question for you .Is it possible to make a script file to fully automate the reconstruction process from photo to stl file ?
use bash scripting, Luke
great overview. thanks a lot!
excellent video and congratulation for the snap package!!! just one question. Why did you multiply the width by 1.2 to obtain the focal length in pixel?
tanks again
Good question! It's the value that the authors of these programs recommend using. I'm not sure how it is calculated. :-)
This is an extremely useful video! Thank you.
What hardware are you using? CPU, RAM, GPU, etc...
He is using Intel P4 processor + 1 GB ram !
this was really helpful thank you so much
Hello i follow your steps to install and it was great! i just had a problem after compiling using blunder it generates many files except for the bundle.out i got the bundle.init.out and bundle.log, but not the bundle.out. could you help me?
I use Visual SFM + Cloudcompared + Meshlab, unfortunately I lack knowledge to do the texturing in meshlab and export image to GIS software.. thanks to your excellent video I think I found the missing pieces
that's a lot of work thank you very much
Very useful overview
thanks for your work and your video very interesting- but i tried to install snap visualsfm-mardy on ubuntu 16.04 and sudo snap install blocked my computer- thanks
Thank you so much for the video :)
great video helped a lot
Thanks for helping me!
Amazing ! Thank you !
Good video, thank u!
This is amazing
What are the hardware requirements to run this type of software?
you saved only one reconstruction from visualsfm, where did the dense image come from??
hello! thank u for your video! it's so nice!!!
so I have some question. can u help me??
I'm trying to use one of these software tools for making 3d model from tree pictures.
what do u recommend to use for my project??
I need to use open source program. please help me thanks a lot!
Trees are very complicated, especially if they have leaves. I would start with mve, but if you are not satisfied with the result you should simply try all of them :-)
oh really? Have you done 3d constuction from nature object like tree, flower..? anyway thank you for answering me! :) I'll start with MVE!
can you make a video on how mve is install
Great video.does anybody knows is there any software from wich i can measure objects from one picture if i know some of distances, but other not. For examplei have picture of building, and i know buildings lenght,is there any way to know hight,lr width and hight of windows?
www.photomodeler.com is a good tool for this type of task.
how we install bundler in window10
very interesting
Hi, I used the same software(same version) and same image data set, the only difference is that I use windows system, but I always failed to reconstruct the dense cloud. Is there anyone same with me ?
already solved this issue, thanks.
Agisoft photoscan is absolutely perfect.. but it’s expensive
Hello!
First, thanks for the snaps and the video! I'm tryng to do some scans, but I get this error when I start visualsfm, and when I try to reconstruct sparse it crash.
ERROR: Unable to write to log file
/snap/visualsfm-mardy/2/bin/log/[17_06_02][13_54_39][132].log
I hope you can help me! thanks!
That message about the log file is harmless. Could it be that you are running out of RAM?
mardy.it I don't think so, I have 14gb
Actualy I have 16 gb of ram... so I think that is not the problem, it crashes in 1 second. I really want it to work!
The good: Good video to get into SfM quickly. Very useful.
The bad: Many of the snap packages from the author (namely Bundler, openMVG, openMVS) did not work on my computer (ubuntu 18.04), segmentation fault, I had to build them myself.
Colmap snap is old and with no CUDA.
Diagram error: openMVG should fill 1st column only, while openMVS fills columns 2-4 (dense cloud onward)
I got segmentation fault for bundler either. How did you solve this problem?
@@hangli1622 This video is ancient. None of these snap packages work on modern Linux. You'll have to install software in other ways, usually building from source.
@@oleksiygrechnyev7120 Thank you.
COLMAP can do CPU and non cuda reconstruction
i found out two commandline options for that. Is there more options?
Can you also include PMVS?
I think I tried it, but for some reason I was unable to make it work.
isn't handled within VisualSfM?
I would love to use this, but it fails in Ubuntu 16.04 LTS (LTS Only here :-) )
For every step I try to fix it fails on another, then back again when I fix the next.
Are you still maintaining this? Getting illegal instruction when attempting to reconstruct sparce, and that was after I blow a hole in the snap protection to get it to start at all....
It looks like a lot of strange compatibility stuff is leaking with the updates nowadays... :-(
I bet that does not help you ...
Well, hoping to see updates that work, I really appreciate your attempts to get into a manageable form...
I also used this on 16.04 LTS only. I haven't touch the snaps since I made this video, but given how snaps embed all their dependencies, it should just work (unless something broke in the snap system itself).
But which snap are you using? It's not clear from your message which programs you are launching :-)
snap list gives:
Name Version Rev Developer Notes
bundler-mardy 0.4-20170208-2 3 mardy -
core 16-2.30 3887 canonical core
meshlab-mardy 2016.12-20170302-3 5 mardy -
visualsfm-mardy 0.5.26-1 2 mardy -
This is the first time I've used any snap on this machine...
visualsfm starts with gui but will not do any processing.
Meshlab simply won't show any gui.
Visualsfm finish feature matching in 0 time, no errors no result.
Sparse cloud step 0> core-dump
You could try to see if the "meshlab" (not "meshlab-mardy") behaves better.
I found a meshlab in the apt-repos. But I can't feed it since everything that create sparse and dense cloud data, core-dumps.
Could this have to do with forced updates related to the Intel bug?
What I get is 'illegal instruction, core dumped'
Lag+big lag+huge lag+gigantic lag
2:54 VisualSFM
9:19 COLMAP
11:32 Bundler
13:05 CMVS
14:20 MVE
19:50 openMVG
BRO! Why didn't you lead with Agisoft Photoscan or reality Capture, these other software you are listing are rubbish!!
Possibly because $$$$$$ ?
3df Zephyr Lite is free. I used the standard version and it is quite good. The only difference is that free can use up to 50 photos max
But Agisoft is not, it's really expensive actually.
It depends,
What do you get for that money?
If you get reduced time spent on processing, higher accuracy and easier dealings with any problems, it might just be worthwhile investment, asuming that this is not for hobby but actual work, where every hour costs money.
The Pro version of AgiSoft Photoscan is $3499.99.