What cameras/lenses are preferable for photoscan?

[ikercito]

If Lightroom has the correct profile for the lens and camera combination the correction will be almost automatically applied. No problems there...

On the other hand (I'm a newbie and just playing by ear) I'm thinking the wider the lens, the more sensitive PS will be to perspective errors and CA. One single frame will contain a lot of perspective information, so the points detected sometimes will be too far a distance from each other to find matches, if there's perspective deviation. Maybe shooting with a slightly longer lens (28? 35?), and thus shooting more frames will probably help getting more matches. Maybe I'm talking nonsense here... Can't help you much, that's simply what came into my mind out of the box. Better get some help from experienced users.

You could try the PS demo and get the camera and lens for testing to give it a try... The software is amazingly capable, so I'll doubt you'll be having too many problems anyway.

[hengefjes]

You could try the PS demo and get the camera and lens for testing to give it a try... The software is amazingly capable, so I'll doubt you'll be having too many problems anyway.

]Well, I'm already using Photoscan, and get nice results with a Sigma 10-20mm f4,0-5,6- lens on a Nikon D90. I use the lens on 10mm (crop factor 1,6). I want to upgrade my gear to get as good results as possible, and since there is no physical photo-gear-stores anywhere around here theres no way for me to try it out. The lens has a pretty hefty pricetag, and I have to make shure it's for me before i purchase it. It is said to be one of the sharpest wide-angled-lenses around, and 114 degrees field of view with good sharpness towards the edges could reduce the number of shots needed in my workflow.

Btw: My e-mails to "info@agisoft.ru" gets rejected by their mailserver-provider

[andy_s]

Hi ikercito, it seems [to me] that you are not talking nonsense.

While this discussion http://www.agisoft.ru/forum/index.php?topic=330.0 didn't end in a particlularly logical conclusion:

"the thesis of the paper is to throw everything [including the kitchen sink] at it"

some comment there may be worth re-visiting.


Also, here's something to be aware of ?

as i'm sure you know http://downloads.agisoft.ru/pdf/photoscan-pro_0_9_0_en.pdf @ page 6:

"...To estimate the field of view for each photo PhotoScan uses the information saved in the EXIF part of each picture. If EXIF data are available you can expect to get the best possible 3D reconstruction. However 3D scene can also be reconstructed in the absence of EXIF data. In this case PhotoScan assumes that the 35mm focal length equivalent equals to 50 mm and tries to align the photos in accordance with this assumption. If the correct focal length value differs significantly from 50 mm, the alignment can give incorrect results or even fail"

http://chdk.setepontos.com/index.php?topic=8613.msg90258#msg90258:

"The EXIF (Tv Av ISO) is not completely accurate in any of these, especially ISO // it's really off sometimes.  For example, I've noticed that if I give shoot -sv=100, the EXIF ISO is 164 and the image is indeed brighter that pressing SHOOT with Canon set to 100.  So there's something weird here too."

"ISO difference is probably "real" vs "market", which I explained at least once before. 100 real -> 164 market is a pretty typical delta. For other parameters, some minor variation between Canon display values and actual values is expected. E.g what displays as 1/125th could really be ~1/128th etc and different values could show up in different places."



PS That whole [long] chdk thread is fascinating - a condensed "War and Peace" e.g the chdk developer in chief says, at one point, in http://chdk.setepontos.com/index.php?topic=8613.msg90969#msg90969:

"I *like* CHDK. I've spent a lot of time on it. It's pretty cool. It would be cool if it's useful for your instrument. All that said, I know the quality of the code, I know much of it was contributed by people with limited programming experience, I know all the crazy unsafe things it does under the hood, and I know that what passes for "QA" is mostly people who don't really know how it's supposed to work trying random stuff on a tiny subset of the supported hardware."

  
 
Edit:
http://chdk.setepontos.com/index.php?topic=9443.msg97933#msg97933:

Placeholder
"Comparing the EXIF data in the JPG and DNG images, I see that the Focal Length is sometimes incorrect in the DNG files.  For example, I took five shots at various zoom levels (from lowest to highest), and I got the following focal lengths recorded in the JPG and DNG files respectively: (5.0, 6.9, 10.7,  24.0, 40.0) and (5.0, 6.9, 10.7, 13.4, 13.4).  The label on the camera states that the focal length goes from 5.0 to 40.0.  I have no idea if this problem is specific to the A4000, or if it is a problem with CHDK in general."

http://chdk.setepontos.com/index.php?topic=8585.msg89924#msg89924:

"...The effect of CHDK overrides on EXIF values isn't very reliable, and varies between cameras and probably camera settings...In some cases, CHDK overrides do not update the regular EXIF, but do update the Canon "maker note" fields..."


Edit:
http://www.agisoft.ru/forum/index.php?topic=330.msg5193#msg5193:

"...The bigger the distance between stereo-pairs, then better the results. With a short basis between pictures the angel for triangulation is small what causes errors(In object direction). I was always using all pictures because I was sure that photoscan is using more Stereo-pairs to estimate one point..."

http://www.agisoft.ru/forum/index.php?topic=330.msg5395#msg5395:

"...Yes, this is generally true, but only to a certain extent. If the distance (viewing angle) between images becomes too large, new problems will appear, mainly related to increasing difficulties of matching features because (a) the changes are too large between images and (b) low incidence angles (i.e. grazing view of objects). In essence, image distance should be neither to small nor too large."

http://www.agisoft.ru/forum/index.php?topic=330.msg5421#msg5421:

"For multiple stereo models, small such measurements can be useful for the validation of measurements with strong intersection geometry. Even though small intersection angles lead to noisy results, models with small base lines should be acquired and used within the surface reconstruction. Since large baseline models have lower image similarity - which is challenging for the matching method, small baseline models are required additionally. Furthermore, highly overlapping imagery leads to high redundancy, which is beneficial for the precision in object space."

[ikercito]

hengefjes: Any specific reason to change your equipment? A D90 with the 10-20 sounds very reasonable to me. You get 2mm wider with the Oly setup, but that's an almost marginal gain. I hope there are other reasons you want to change your gear...

On the other hand you say you want to minimize the number of shots on your workflow, and I guess that's not a clever idea. You can try to squeeze as much info on a single frame, but remember that each shot is going to be one single persective of your scene, and from my little knowledge in PS, the more points of view you provide... the better the scan will be. And also, the wider the lens, the bigger the perspective errors and CA will be... and that will cause getting fewer matching points between shots. So as a rule of thumb I'd say the more frames you provide PS, the more info it can extract from your images. Perhaps I m mistaken... but that's what I found out worked best in my scans.

Andy_s: Regarding EXIF data, unfortunately there's always some deviation. I don't think it matters much with exposure values, you get what you get in the picture whether it says 1/125 or 1/142, same for aperture. ISO is known to be way off in most cameras, specially for marketing purposes (noise, sensitivity...). But with focal lenght that can be more troublesome, since PS relies heavily on that info. I'm sure there is some margin where PS can adapt to those inconsistencies...

[hengefjes]

You can try to squeeze as much info on a single frame, but remember that each shot is going to be one single persective of your scene, and from my little knowledge in PS, the more points of view you provide... the better the scan will be. And also, the wider the lens, the bigger the perspective errors and CA will be... and that will cause getting fewer matching points between shots. So as a rule of thumb I'd say the more frames you provide PS, the more info it can extract from your images.

My idea was that the microfourthirds would provide photoscan with more usable info towards the edges of the pics, since it's supposed to be very sharp and already fixed by the systems built in lens correction. These m43-systems are also said to have excellent depth of field.

Our workers are taking pictures constantly while restoring boats/ships, and we have to get more cameras anyways to make sure they are available when needed. My bosses are very impressed with what Photoscan gives us, and wants me to find a good camera for it - and my other photo-needs. We thought 48 gigs of ram would be sufficient for our needs, but I guess I have to squeeze in the last 16 gigs of ram that the motherboard supports.

Would it be a better approach to increase the focal-length and number of shots, but at a lower resolution - so that the computer can cope with it?

Added: Heres a link to my "Sketchfab-page", where you can see some typical scenarios. I haven't done many interiors, as i need a good lighting-system. Would really apprechiate some suggestions there too. I'm new to this too...
https://sketchfab.com/gunnar/recent

[chadfx]

Super wide angle lenses will by nature have a fairly deep depth of field, so I'm not sure if you will get a whole lot of improvement with micro4/3 in that regard. Are you finding that you are seeing DOF softness issues with your Nikon setup? Maybe if you have some foreground objects that are very close to camera that it could be an issue, and possibly for some of those tight interior spaces that sort of issue would come up.

Although the other of sharpness issues at the edges of frame are certainly problematic with many super wide angle lenses, so finding the ones that deal with it best could be worth exploring. Also, a fixed focal length lens could possibly yield better results than the zooms you have been using and/or considering, as they generally involve fewer optical compromises. (not always the case, but certainly worth keeping in mind)

If you could find a place that rents lenses online, you might be able to try one out before you commit to buying. Unfortunately the online rental companies here in the US don't ship internationally. (lensrentals.com & borrowlenses.com)

[JMR]

Hi, hengefjes
I have that lovely lens (zd7-14). You'll be amazed by its DOF . But I must admit i have no tried it with pscan. Ill take a bunch of photos and share them with you asap.
Btw, have you considered the same 7-14 by panaleica an available in mft mount? I have used it sometimes an its pretty good also. (It is not weather sealed nor that superbly well built as Zuiko, but is better balanced in a micro fourthirds body.

[andy_s]

Found something... I'm not sure if the diaphragm column refers specifically to having a real aperture... But i guess it's a start

http://chdk.wikia.com/wiki/CameraFeatureTable

SX160IS has Manual Focus and aperture

http://chdk.setepontos.com/index.php?topic=9443.msg102555#msg102555

Quote from: reyalp on 06 / July / 2013, 22:37:08
One exception: If the camera has an MF mode, set_focus() should take immediate effect.

That's good to know.  I'll try using my sucky method for setting the focus, but if it get's too frustrating, I'll buy a couple of cameras with support for manual focusing.  The SX150/160 are pretty cheap these days.

[not yet sure about external power]

[hengefjes]

Chadfx: I suspect I have DOF-issues when working in engine compartments, as tubes, rails and other "thin" stuff in the middle of the room seems to "blur away" in the model. I know i need better lighting, and should probably pay more attention to texturizing shiny surfaces. I've been using chalk and colour-powder mixed in water, sprayed on and left to dry before shooting - which seems to help, but this is not a good solution either since these rooms are very narrow and I need to be able to move around while shooting. I also suspect I may have used too high aperture in my tests. - Unfortunately there's no rental companies here in Norway, that carries the gear I want to try.

I've been considering the Nikon D5200 because of it's flip-screen and focus abilities and a Tokina 11-16 f/2,8, which gets very good reviews. The Olympuses also have flip-screens, and is said to have very reliable focus.

JMR: That's such a GREAT offer! Would it be too much to ask if you could take some shots of the lens-calibration patterns too  I've seen that Lumix G Vario 7-14 online, but got the impression that its lens-calibration would only work on Panasonic cameras. I'm not familiar with mft cameras, never tried one either. Wifi could come in handy. Do you have any suggestions for what camerabodies i should consider? OM-D EM-1 and 5, and the EP5 all seems great - bit pricey though, as we would have to afford that lens too  The Zuiko is about 20% more expensive than the Lumix 7-14 - is it worth the extra money?

[chadfx]

Hengefjes: There's a fairly reliable way you can predict DOF issues...use a DOF calculator.

http://www.dofmaster.com/dofjs.html

As you can see in the examples below, with a 7mm lens on micro 4/3, even wide open at f4, you can pretty much get everything in focus unless it's right in front of the lens. But you can also get fairly similar results with your Nikon @ 10mm & f4.

If you had shot with the lens stopped down too much (ie. f22 on your APS-C Nikon), then yes, you might have wound up with everything in the frame just a bit soft. Those thin and shiny objects will likely always be somewhat problematic, and good lighting and exposure will certainly help overall.

Don't forget that if you are shooting RAW images (of course you are!), that you can tone down the highlights and push up the shadow areas quite a bit, which can help to avoid noisy geometry reconstructions. It might be worth trying to 'redevelop' your images from a previous shoot and see if you can get a cleaner Photoscan build from them.

One thing about the Olympus OMD cameras (and the EP-5) is that 5-axis stabilizer they have. The IBIS works on any lens (most super wide angle lenses aren't stabilized), and can help you deal with shooting in small dark places where you might not be able to set up more lights or use a tripod.

[hengefjes]

Chadfx: So far I've done it all directly from jpgs, but I'll certainly dive into Lightroom hereafter. I'm pretty new to photography and photoscan, and learning as I go, - and just now I learned a whole lot from your link. I've been around f16+ - I'm always using a stand anyways. Someone told me - the more aperture the sharper, but reading all those lens tests made me suspect there might be a sweetspot. Great info!

I'll stick with the Sigma 10-20 for now - or at least until I have learned more, but the D90 has to serve among the craftsmen - I tried a D5100, and found that flip-screen to be very handy - also the focus, light measurement and whitebalance seemed to be better than the D90.

[chadfx]

Yeah, I know what you mean about it all being a lot to learn all at once. Your examples are looking pretty good so far! That suggestion about smaller apertures yielding greater DOF (and sharper photos front to back) is 'generally' true, except that there is a phenomenon known as diffraction which starts to occur at very small aperture openings and actually starts to make the image softer overall (even the parts right at your focus point). Sometimes people will talk about a point where a lens/camera combo becomes 'diffraction limited'. That point usually comes up faster if you are dealing with a smaller sensor size. So for micro 4/3 cameras, a lot of lenses will start to show a bit of softness once you go past f8 or so.

Here's a review of the Panasonic 7-14 with a great little web gizmo that lets you slide through the aperture settings and see the softness values start to decrease when stopping down the aperture, and then increase quite dramatically when stopping down even further:

http://slrgear.com/reviews/showproduct.php/product/1252/cat/69

I would agree that you will be fine sticking with your Sigma 10-20 and Nikon combo until you really find you are at the limits of what you can get with it. Shooting with RAW and other efforts like lighting and comprehensive photo surveying will probably give you results you (and your bank account) are happier with. I am also a big fan of the flip screens and live view can be really handy for fine tuning your focusing (and with live depth of field preview, you can get a very good idea of how much of your image is in focus).

Cheers, -C

[ikercito]

Hey hengefjes, I'm a professional photographer and photography teacher...! Need a Photography Basics 101?  Hahaha! just joking... but any help you need, just ask. Everyone at the forum we'll do our best to help you out.

BTW your scans look very good. Congrats on those. I'm sure they can be improved, but for now, I'd just stick to the equipment you have and try to learn the techniques involved in photography and photogrammetry, and try to squeeze your setup as much as you can. It happens very often that people think they need better (or different) equipment to get better results in anything they're doing, and most of the times they are far from using their own equipment to the limit of its capabilities. It's the most common syndrome in photography.

On another topic... I just got a Kinect for Xbox, and been playing around other softwares to scan stuff. Unfortunately the quality is quite bad for scans, but that machine sure has got possibilities in motion capture! I'm really having fun...!

[hengefjes]

Ikercito:  I was also playing around with the Kinect a while ago, and hopefully the drivers are updated now. There where to sets of drivers which I had to switch between to use different software with it - almost drove me nuts   Got some nice results with a program from Faro (Scanect og Scenect), and some with the toolkit that Microsoft provided. To let the sofware track the movements I had to move veeery sloooow. There's a new Kinect coming with better resolution from what i understand, but i guess the range will be about the same - after all its just a toy, but interesting technology though.

We have tried laserscanning here at the wharf, but the prices for the equipnet and software makes that stuff utopic. From what I can tell, Photoscan/photogrammetry even does a better job in our case. I tried to compare meshes from the two in CloudCompare, and the model from Photoscan agrees with the laser where the laser has been scanning directly towards the hull. The lasers precicion seems to decrease with the angle towareds the object  - i'll ad some illustrations. - The conditions were horrible for the laser though. Lots of scaffolding and tight space around the boat, and as you can see in the pics, the ones that did the scanning haven't been able to register the pointclouds properly either.

Added: With photogrammetry we can do the geometric documentation whenever it fits us, and not under horrible conditions because the appointment with some company says so. - Also we are supposed to take pictures of everything anyways while we are doing these restorations.

Since I'm already way off topic. Is there any cameras that write tilt- and compass-data as exif info? Couldn't that have been useful in photogrammetry?

[ikercito]

Hi again guys. I've just completed the first test with my 4 camera setup. It's nothing fancy, just a cheap rig with a bunch of Canon point and shoots (two Ixus 105, one Ixus 220HS and one Ixus 85). I've set them up on each corner of a wooden square frame 40 or 50cm appart from each other, connected to a usb hub and shot thru CHDK Remote. It works fairly well and all cameras sync with an almost unperceptible delay. Just switch power on to the hub, cameras focus, switch off power and there you go... cameras shoot with no apparent delay. Looks fine...

But I must admit that I'm quite unimpressed with the results, they're just not what I expected. For now, I've had much better quality scans shooting with one single (and better) camera around my subject, around 20 to 25 shots while my friend stood still... Than the quality out of 4 synced shots with these cheapos. I believe they simply don't have enough resolving power (sharpness) for this task. Obviously this first test comes from only 4 shots, which is just the minimum... But setting up a bigger rig is out of the way right now. Not with these results! My next test will be using both approaches, shooting a few rounds with the rig around the subject in order to get 20 to 25 shots. They won't be synced anymore, but the time to get it done will be much shorter. No more messing around with one single camera... Now every click will be 4 shots at once!

And one more con... It is a real pain to set CHDK on each camera, and most of all... once all shots are done, the worst part is downloading everything to the computer. With CHDK remote on, you cannot access the camera thru usb, so every card needs to go out of the camera and on to a card reader.

Will keep testing to see how to improve the workflow, but  for now those are my first impressions.