Lukas Mach is developing software that converts a series of photographs into a textured 3D model.
From the website:
insight3d lets you create 3D models from photographs. You give it a series of photos of a real scene (e.g., of a building), it automatically matches them and then calculates positions in space from which each photo has been taken (plus camera’s optical parameters) along with a 3D pointcloud of the scene. You can then use insight3d’s modeling tools to create textured polygonal model.
Basically, it hopes to be "Photosynth + Canoma". Or "boujou + Imagemodeler". But it's also free and opensource (GNU AGPL 3).
The software is still in development (isn't it always?) and Lukas needs your input to tweak it further:
But I need help: there are tons of different digital cameras and I've tested my app only on few of them. insight3d should be able to handle all typical cameras, but there's some stuff in the optimization routines that has yet to be tweaked appropriately. Probably. It would be great if you could send me datasets you tried to feed into insight3d. Both the ones that worked and the ones that didn't are imporant! If you want, download the app, read the tutorial (there's important info on what photos should be used and where to eventually send them!), take some photos, have fun with it and please send it to me. Hope that it'll work for you, but don't hold your breath.
The software is available for Windows and Linux. No OSX (yet?), so I can't tell you how well it works. If you guys could post some sample models that would be awesome!
Link
35 Comments
Good to see that this project is still going. I briefly tried it out a year+ ago, and remember it working well.
can it be co-used with blender ? can the code can be ported for blender ? i love to dream to get these features inside our software...
@Nathaniel: a year+ ago, did it look like now or is it actively developed..?
I just tried it and it is actually very well made. Couple of UI tweaks here and there and it would be perfect for OSS archViz.
I've been looking for this, I was happy, (no OSX), then sad.
I love to dream and get these features inside software, i can blender, and code ported
From the insight3D website:
Disclaimer: This software is intended for educational purposes.
This means no prof use?
Nice stuff, will take a look at it and if it could replace some tools in my pipeline.
Given the license, while he might intend for the software to be for educational services, the AGPL 3 doesn't allow for that restriction. If it released under the license mentioned above, it will be for any purposes the user decides so long as the program is distributed under the GPL :)
There is a dependency missing in the README.txt file. You should install libhighgui-dev.
Can anyone compile it? I'm on 64bit Ubuntu 10.04 and i'm getting this error:
g++ -O3 -c `pkg-config --cflags opencv libxml-2.0 sdl gtk+-2.0` -I./ann_1.1.1/include/ actions.cpp
In file included from geometry_structures.h:38,
from actions.h:28,
from actions.cpp:25:
./sift/include/utils.h:159: error: new declaration ‘char* basename(const char*)’
/usr/include/string.h:601: error: ambiguates old declaration ‘const char* basename(const char*)’
actions.cpp: In function ‘bool action_camera_resection(size_t, bool, bool)’:
actions.cpp:108: warning: format ‘%d’ expects type ‘int’, but argument 4 has type ‘size_t’
make: *** [actions.o] Error 1
A GPL3 project "for educational purposes only" to me means that the guy is doing it for fun, ie for his own education. He is not intending to make this his day-job, just a hobby. One should not try to do (semi)-proffesional stuff with this and expect the writer to fix things in a timely way.
Of course being GPL3, you're perfectly free to use it for whatever use you want (educational, proffessional, perfect it and use it as an in-house tool etc etc)
The way I read the remark about educational use is that it's only to protect himself from patent claims or from professional users who screw up a project with his software..
Lovely!
Rio, you have to copy the 32bits libs to ..patch/insigh3d, then you have to set the patch to system: export LD_LIBRARY_PATH="..path/insight3d"
I tried to build insight3d and didnt have sucess, but using this tip I had sucess with 32 bits precompiled executable.
Nooice!
@Rio : I managed to compile it on ubuntu 10.10 64bits after a lot of trials and errors (check here if you are interested : http://ubuntuforums.org/showthread.php?t=1609442&highlight=insight3d). Unfortunately, it seems it is not usable yet on linux, it crashes everytime I tried to compute the 3d model out of the photos.
I'm not sure it's still been developped.
Seriously cool! Would be great if it was integrated with Blender.
I did manage to compile it on Ubuntu Lucid amd64, but it always segfaults at calibration step
Wow! now that is something different just from images you bring out a model of it whoa i'd like to try it great work Lukas keep improving it.
sounds very interesting. Can't wait to hear more. Blender-integreation would be insanely great.
great to hear that this project is still runnin:)
makes me dream of endless possebilities... tracking/matchmoving... generate 3d mesh or panorama out of movie sequences..
...you dont have to model your backround of your tracked clip, its already done and textured:))))
Still 0.3.2? As my .msi file data says I must installed it last time in August this year then at XP not actuall Win7, but had no sucess. Any subjects which are going to give best or any results?
ubuntu 10.10 64bits works for me. As far as getting it to blender vmrl doesnt seem to work right. No textures. I found that libmv is getting up to speed and have been doing testing with that. Libmv get a nice point cloud and camera via a py script it creates for 2.49. Havent done extensive testing. Found that I was able to get a camera track by copying and pasting the solved camera positions and rotations to the main camera. Thought that I would take it a step further and bake on the texture using uv project modifer having lots of crashes havent figured that work flow out yet. TerryWallWork sent me a nice tutorial on how to bake uv projection mod to a object but I havent got it working yet....
I'm using Ubuntu (10.10) and I couldn't figure out how to compile and run the native linux version, but the .exe ran perfect under Wine!
This is definitely the type of program that would work best to create architectural mock-ups. And with my own tests it is definitely important to have good input data, and even with good input it still requires a lot of tweaking to get a good model out of it. The automatic only goes so far.
I'm still excited to have this as another tool in the 3d belt!
No second thought wine is what was using and worked fine sorry
I checked out the software and it seems this software could use a wizard guiding a user through the steps. Using the demo images and the user guide I came to the conclusion that it's a lot less automated as I first thought. I thought this app would import some images, it works its magic by matching the images and calculating the perspective on it's own. You know, kinda like that modeling tool that you rotate some object in front of a webcam and it generates a textured mesh.
Finally I came to the conclusion its much less work to model the building from scratch within Blender and slap on an UV Texture. I don't know, but the GUI needs to be more too.
Maybe it's just me :)
IMO this kind of app is not about modelling and texturing in one go.
Blender can do that. The important thing is the ease with which you can model very precisely from photos.
Many older buildings for example has a very wierd geometry and its very hard measuring that in real life if your not running around the building with a measuring band and even then its hard to model it precisely.
This kind of program is really interesting for all architects, because of the precise model. The texture is just an added plus. :)
I think it was Karpov or Kasparov who said that after he lost a chess game against the computer.
-The racing against the computer like racing in the weightlifting Olympic against a forklift. ;)
I think the model copying is getting somewhere here. So be creative if you ..
When you export a camera as .txt, how can it be used in Blender?
If only paint projection had (tracking or detection) in Blender.
You could take the point cloud from(libmv) convert it to a mesh(2.49) and uv projection paint onto it.
Nice! I want todo a importer for Blender. How works the "cameras".txt file? I can`t understand!
Where we find the source code? It would be great to make a port to python to integrate it in Blender.
Tight the phone to the camera. The shooting voice is recording on the phone (for time).
After the shooting (photos), Send the voice on sms.
And the phone service gonna send back the chain positions of the camera.
What do you think? They love money don't they?
I looked at the project, because the program looks promising, but:
1. The development dosn't look very active:
- only one SVN check-in
- only one developer
- no applied patches
- no response to bugreports
- the developer doesn't response to mails
2. Not very Linux friendly:
- exe, dll files in the Linux tgz
- no configure script
- dosn't compile on Ubuntu 10.10
3. No MacOS X support
4. No Licence file in the tgz
I've done it! I created a fork. Not That i love forking, but it was the only way I had, because the original author did not response.
You can find the fork here:
https://sourceforge.net/projects/insight3dng/
Every one in invited to help with this project.
Changes between insight3dng and insight3d:
- applied patches
– no exe, dll, *.o files in the tgz
- compiles on Ubuntu 10.10
- new git repro, open for every one
Hi,
I have a problem with import images into the program. When I enter the picture only appears white image. Do you know what is wrong?
Tnx