Suggestions on how to get better

Hello everyone,
I’m trying Reality Capture these days to understand its possibilities in my field of work (architecture).
As such, I’m trying to capture room and apartments to get better at it but while I’m positively impressed by the general speed of the program, I’m having some issues with the alignment phase and the subsequent reconstruction phase, in which I cannot produce models without holes, even if there are no reflections or windows.
I tried both with a reflex camera and an iphone (just to try different layouts); I think I may be taking lousy photos, for example I can never get a surface from a room’s ceiling!

I’m posting a photo, typical of the situations I meet during my experiments !
Can anyone suggest some rules I should follow, texts to read or just tips to enhance my abilities? : )

Thank you in advance

Hello Giorgio,
some tips for you:
Taking pictures for photogrammetry
Yellow triangle with an exclamation mark
Calibration groups

Hi Giorgio
Interior rooms are the hardest subject to do… So you need a lot of experience with photogrammetry and it is better to use a laser scanner if possible ( can be rented for a good price on per project basis )
How many images do you have there ? what camera and settings did you use when taking pictures ?

Hi Lubenko,
thank you for your kind suggestions, having read those articles I now feel I’ve got several things to improve : )

Hi Wishgranter,
I’ve got about 150 photos for the room posted before, should I try with more? I’m using a nikon D5100 + 18-105mm 1:3.5-5.6 is it enough to get decent results? Can’t access to the actual files now, I just remember I was following the general tips on the camera use to shoot !

Thank you,

Hi Giorgio

Yup, much more… try to get to approx 300-500 images… multiple angles ( good for model quality ), a lot of overlap and SHARP low ISO…

I’ve been doing a lot of inside photogrammetry lately - it gets a ton easier as you practice! I’ve been able to capture some pretty fun scenes just with some handheld shooting, but I invested heavily in a camera with IBIS (a6500) and a really good wide angle lens (Zeiss 12MM). I have been surprised how much better my shooting got after the upgrade, but I was still doing ‘well’ with an older camera and a cruddier lens. Shooting at 11 fstop is also heavily recommended.

Take tons of pictures, even if you hit 1 or 2K you’ll be fine. Regardless of good practice you read priority one is no blur, followed by 100 ISO or lower whenever possible (I used to hit 200 or 400 ISO to keep some exposures down below 2 seconds in low light, but I find staying at 100 ISO is best regardless of the exposure length). I use Image Stabilization as for those long exposures even on a tripod as people walking around (such as myself) can give some shake and add that blur which tanks the scene.

Once you get the hang of it, you can burn through indoor scenes rather quickly depending on what you’re hoping for the end result. What has made me improve in leaps and bounds is I spend 15-30 minutes max during a lunch break to see if I can capture something handheld every other day. Tons of fun! Nothing better than going to bed, putting your lunch time scene in the queue and waking up to see how well it captured. Each iteration you’ll have a light go off in your head and your next scene collection will be that much better.

These have been decimated for the web:
Somewhat outside scene in some solid drizzle, 15 minutes:
20 minute shot at a friend’s store:
(better detail of a table):
My first indoor shot, done with some crappy hardware: (started learning my post 3d work in this one so it is midway first attempt and looks real bad as a result)

On that last one, I went in to redo a few exhibits with my new equipment and realized after I RCed it that I could have done the whole room in literally 1/3rd the time with much better results with what I had learned and my new set-up. So stick with it!

Hey Castlenock,

I see that you’re back in the forum!

I’m trying desparately to decide on a new camera system but can’t get my head around wheter to go for something with stabilization or without. Almost anywhere with professional advice they say it’s a big NO NO. But then I know of a few people who use e.g. an A7Rii win IBIS and they have no problems.

At the beginning of this post you also recommend to switch IS off, but in the last post you said you bought something with IBIS and use it a lot. Do you have any idea why IS is supposed to be bad for photogrammetry? Is it possible that this is mainly true for software that needs the same distortion model throughout, whereas RC does it on an each image basis anyway, so it doesn’t matter so much?

From my experience with my older canon 6D setup was that lens Is did cause very noticeable issues.
However it was on zoom lens’s so not really ideal for the setup.

With the a7rii the IS is better when combined with a good compatible lens.

It does and can cause issues, anything that manipulates the images can make hard work for the software and cause issues.

That said the advantage you get from IS means you can get sharper pictures, or as castle states a lower iso (less noise) from the same exposure time, or allows you to reduce the exposure time, which really helps when hand held.

So it’s a trade off between the two.

The software works best with sharp low iso images,  but doesn’t like IS.

If you turn off IS your more likely to have to turn up iso (more noise) and/or have a longer exposure time (more blur).

Which is best ?

  • It’s situational, lense, available light, handheld/tripod all make a difference and something that I do aim to do tests with and provide datasets for (benchmarking V2) that won’t be today :slight_smile:

The A7rii Is a awesome camera, it’s expensive and so is good glass.  It’s not as nice to use as a cannon, but miles nicer to use than my sigma dp0 (foveron) which does incredibly accurate shots - perhaps better than bayer cameras… for this application, but is not user friendly at all.







Hey ivan,

thanks for elaborating.

The thing is that I would like to know WHY it is supposed to be so bad. Is it because the center of the lens shifts?

That should not be too much of a deal in RC due to the individual distortion of each image, right?

Also, is this not mainly important for real metric cameras? Any consumer grade camera, and the A7 and all “normal” reflex cameras has issues (in terms of internal stability) that knock the top of accuracy right off. So it’s more of an intellectual difference, if I read your thoughts correctly.

There have been issues with the A7ii series, but in many cases the IBIS could be ruled out. Many other users are totally pleased with them.

Fear my paint skills !

The camera has 5 axis stabilisation and other factors such is not having a global shutter - (no consumer accessible ones do) come into play. maybe movement in multi axis causes more distortion.  -  I do recall seeing some accuracy tests done with on/off It was interesting, and different situations caused issues at different times.  - It is pixel peeping - however the software does love pixels…

Here is my totally unscientific and inaccurate & exaggerated portrayal of how things work.  Although I believe fundamentally correct… I think…

Based on regular awkward, non ideal lighting conditions.

Hey ivan,

if I am supposed to fear that, then I think I like being afraid!  :smiley:

Awesome illustration! Very clear and intuitive. Add DOF to it (as in focus issue) and it is a HOW TO for image taking.

Actually, I thought of opening a thread about that, since it still seems to be something that many people don’t think about - in the same way as most people focus on the camera body (pun intended) and not the lens.

But just now, I want to discuss the image stabilizer issue a bit more. I did a bit more reading on the matter and it seems to be that IF the theory would apply to all situations in the real world, there should be no problems. However, this is unfortunately not so. One of the main problems, I gather, is the fact that the stabilizer does things in a way and at a time that we cannot predict. That means it might be too slow at times or over-responsive at others, which leads to an exacerbation of the problem instead of the intended amelioration (yes, I googled the last word). So we have the fearsome black box. What bothers me still ist that there is actually nothing to quantify anything, but maybe that is because there are just way too many possible combinations - including body, lens, object, light and also, very important, the photographer!!

A few things that seem to be worth pointing out is that an IS can’t predict the future. That means it can only respond to predictable vibrations (as in hand tremors) or such. It cannot 100% prevent a shake from singular event as, say, pressing the release button. I wasn’t aware of that but it kind of makes sense since even if it is super fast it still needs some time to respond which means there will be some blurr even if it manages to catch it before the shutter closes again. Also, it can happen that a certain vibration hits the natural resonance of the IS system which means that it will build up a wild wobble. The latter is often a problem in UAVs, which isn’t our main point here (I guess).

Long story short: It can help, but it can also make things worse. There is no one right answer…