Is the following camera and lens good for photogrammetry ? Is anyone here even into this ?
Canon EOS 1300D
EF-S 18-55mm IS
Is this good for photogrammetry ?
[EXIF data available. Click here to show/hide.]
Camera-Specific Properties: Camera Software Greenshot Image-Specific Properties:
No, you need a prime lens. It makes no sense to purchase a zoom lens for photogrammetry, especially one without a lock. Zoom lenses have precisely the wrong properties for photogrammetry, and cause processing to fail if you change the focal length accidentally, as the software has to assume that the focal length is constant in order to perform the coordinate mapping.
Nothing particularily wrong with the camera, but there's better ones to buy with photogrammetry in mind.
The best lenses for photogrammetry are primes between 35-65mm equivalent, (So on APS-C like the cannon you suggested, get something like the the EF-S 24mm, a very, very good lens.
Because generally you want to have the camera at a lower stop like f4-8, as short depth of field is poison to photogrammetry, it doesn't matter what the maximum aperture is. In fact, the best possible lens is a stabilized prime to allow you to handhold in worse lighting while still using the very necessary small aperture.
If you're willing to buy second hand, there are many many better choices.
Generally dynamic range and high iso isn't a huge issue, so buying canon is actually OK in this context. I would absolutely recommend that you shoot raw though. Also, you must not use auto settings. You have to use manual so the camera doesn't change exposure while you're taking your shots.
So yeah, I'll give good combinations per manufacturer / setup
Canon APS-C: anything 550D+ or 60D+ have sufficient detail for most photogrammetry, then use EF-S 24mm which is very cheap and has very good characteristics for photogrammetry.
Canon FF: Get a used 6D + EF 35mm f2.0 IS OR EF 40mm f2.8 which is cheaper and super sharp from the get go.
You can look up the Nikon equivalents if you want.
I wouldn't reccomend Fuji, I have the X-e2 and I long for when I did photogrammetry with my canon. If you do choose mirrorless, the X-e2 is the best choice for photogrammetry, and then get the Fujinon 27mm f2.8, or get the Sony A7 second hand with whatever lens is sharp in the range, but not a zoom.
You can do photogrammetry with a fucking cell phone. Just make sure the lighting and white balance doesn't change. Manual mode.
>>2863081
thanks, men
>>2863082
Jpeg compression fucks photogrammetry so hard, especially adaptive compression. You have to be willing to shoot raw, or halve the resolution of your pictures every time you import them to compensate from the loss of spatial resolution.
>>2863085
Not really. Just make sure you use fine quality. Of course if you use something like 99% compression it's going to get fucked up but I did completely fine with Samsung S2 and Agisoft.
>>2863086
That's what I'm doing. Photogrammetry based games is going to be the next huge thing. You don't need a team of artists if you can just go around taking pictures of the real world and letting your computer process them into assets overnight.
>>2863089
If you really think Photogrammetry doesn't require cleanup and a lot of work outside just taking snapshots of rocks and trees you're a little deluded.
>>2863089
this.
http://steamcommunity.com/id/slastraf
some content of photogrammy by me ,
also browse latest screens if interested and add me if you want.
>>2863093
nvm, didnt upload any screenshots of my latest work with photogrammetry yet. but I will
If you're searching for software or tutorials read below. I had written this for /3/ quite a while back.
If there are any newer open source or free software solutions for this, please share.
The current it tool is VisualSFM: http://ccwu.me/vsfm/
It uses SIFT for feature matching to then solve camera position. After that it can use CMVS/PMVS2, CMP-MVS or SURE for dense point cloud reconstruction.
Point cloud data can be imported into Meshlab to create mesh and texture. Mesh can be exported to obj. along with texture.
The workflow using CMVS/PMVS2 is explained in this article: http://www.academia.edu/3649828/Generating_a_Photogrammetric_model_using_VisualSFM_and_post-processing_with_Meshlab
Video explanation for the workflow: https://www.youtube.com/watch?v=Ax6gajFE-qw
Guide to installing VisualSFM: http://ccwu.me/vsfm/install.html
VSFM feature explanation: http://ccwu.me/vsfm/doc.html
Meshlab: http://meshlab.sourceforge.net/
Potentially better results can be achieved using CMP-MVS.
It requires a CUDA gpu. Because emulating doesn't work I can't confirm how comparatively good the results are.
Video explanation for the workflow: https://www.youtube.com/watch?v=Yt9MmQHobTI
2nd (newer) Video explanation for the workflow: https://www.youtube.com/watch?v=ZRTEMKS3Sw0
CMPMVS: http://ptak.felk.cvut.cz/sfmservice/websfm.pl?menu=cmpmvs
CloudCompare: http://www.danielgm.net/cc/
Photo arrays for testing can be downloaded from Microsoft's Photosynth website using Photosynth Toolkit: http://www.visual-experiments.com/demos/photosynthtoolkit
Photogrammetry presentation, long: http://video-jsoe.ucsd.edu/asx/AguerayArcas.asx
Short: http://www.youtube.com/watch?v=6cnP8FqRoPI
>>2863095
Or just use fucking pirated Agisoft and get better results, faster and easier.
>>2863096
Op here again.
I have heard 50 mm lenses are better for 90% of work with photogrammetry,
>>Canon EF 50mm 1:1.8 II
is this the right lense without zoom ?
Does it fit on the Canon EOS 1200D?
have decided to buy the
Canon EOS 1200D because of attainability ,
germanyfaggs only provide very shitty used content
>>2863101
its field of view is too small on crop camera like canon 1200d. That can be a problem if your object is placed tight with no big room around it. Also it will requie more shots to capture all of your object. So i'll recomend you something like 24 or 28mm IS
>>2863133
idk if you still on this thread,
https://www.amazon.de/Canon-EF-40mm-Objektiv-Filtergewinde/dp/B0089SWZDU/ref=sr_1_10?s=ce-de&ie=UTF8&qid=1465928360&sr=1-10&keywords=EF-S+28mm
this or
https://www.amazon.de/Canon-EF-S-24-STM-Objektive/dp/B00NLBGD1A/ref=sr_1_1?s=ce-de&ie=UTF8&qid=1465928415&sr=1-1&keywords=EF-S+28mm
this
>>2863096
No one asked your opinion, you low energy scrub.
>>2863142
24 is more versatile so this
>>2863151
If you want to spend your energy on pointless things while achieving less, by all means, enjoy. I did try and go the FOSS route, and did achieve a 3D model but Agisoft did a much better job with much less effort.
>>2863091
I've been doing it for months. It's all about technique, once you've got the experience and computing power to spare you can reduce the amount of human interaction required for the software and batch organize stuff to run overnight. As long as every picture's pin sharp and free from CA or movement you can generally trust Photoscan to put out results which are great, just start with a higher resolution than you expect to use and only downsample for in-game polycount. In the process this reduces artifacts, but allows you to go back and rework the source file later. My advice to OP is take it slow - it's always better to have slightly fewer pictures which are all sharp because you took your time, as opposed to many pictures most of which you have to disable because they're actively harmful to the algorithms.
>>2863204
let him smelt metals to create cpu's and write his own operating systems
Any advice on generating 3D models from macro images? I've been having not a great time doing.
>>2863331
Op here , thanks for advice.
I have made some 3d models already for a steam workshop, it got 200 downloads no bad reviews at all. I just needed advice on the camera. Yes , actually if you are just starting get agisoft and just set the polycount to 1k-5k depending on how deatiled. the texures make it look realistic anyways at 500 polys even.
>>2863657
The other reason why I say to keep the high density point clouds and/or models is that you should use them to bake normals. If you use a good decimation form, you can reduce poly counts drastically and rely on normals for most fine detail.
The other thing I do which I'm not sure many others do yet is use photogrammetry to generate textures, especially ones I want to use displacement mapping on, or parallax mapping.
>>2863705
Seamless textures? Can you do that? Would be pretty nice.
>>2863331
how do you do low poly models with pbr then from photogrammetry? You still need to do the retopology, dont you? The automated process gives you very bad results (zbrush zremesher is getting better but still shit for high end product)
>>2863726
Manual retopo is always best, but often for organic objects with uniform distribution of detail you can just rely on decimate modifiers .etc
>>2863708
Well, they're not automatically seamless. you still need to go into gimp or whatever and use the heal tool on the edges. Great thing is for flat surfaces you only need to take 10 or so pictures. I often walk in a circle around a clear area I want as a texture, pointing my camera at the same central position. From there you use the bounds in agisoft to create a square portion up to the bounds of the mapped circle. I've been very successful with this but I don't have more examples because I'm away from my desktop. Here's a mud overlay I did a while back.
This is absolutely the future of all real time photo realistic graphics, and the best thing is that it goes hand in hand with VR.
http://www.gdcvault.com/play/1023272/Photogrammetry-and-Star-Wars-Battlefront
http://www.slideshare.net/DICEStudio/photogrammetry-and-star-wars-battlefront
Here you go OP, dudes at DICE use 6d & 24mm (24-70mm L series), take photos in ambient lighting (aka cloudy days), remove highlights and shadows in RAW. For PBR you'll have to try to get the final albedo to be as drab and sad as possible. Different engines will have slightly different smoothness values associated with different materials but if you search for it there should be numerical values (grayscale) associated with different materials on your engine of choice.
