Hi Guys,
So I’m using the demo version and had great results so far. I’m doing my biggest model yet and its 450+ TIFFs. Now it works fine on normal quality but High Detail gets stuck in the same place each time.
I say gets stuck after reading the forums it may just have gotten super slow. I did get a cache disk full issue once but have switched from my SSD to my storage HDD for this one. The same still seems to be happening.
I’ve left it over night and didnt seem to move. I’ve come back to find my computer off a couple of times. Re-ran it a few times after changing the cache but it doesn’t seem to have fixed it.
This is the last test I wanted to do before I bought and moved on to the next stage of retop. Any help would be great.
Any more details you need from me just let me know and i’ll be happy to supply
thanks
4770k
16gb ram
GTX 1070
You mention that you found your computer off a couple of times.
If that is not normal behaviour, have you checked the windows log files for any indication why it shut down ?
It could be heat or power related.
Also the temp file for capturing reality can get full of junk, - try emptying that too. Should be found at C:/users/Yourusername/appdata/local/temp/RealityCapture - just delete the whole folder.
Yeah, I cant see any crash dumps. I use whoCrashed as I had some loose wires a few months ago. Also can’t see anything in windows event viewer about it crashing. No ‘…shut down unexpectedly…’ screens either. I haven’t had any issues with overheating with this computer before and it’s been left rendering for days a few times. No hardware changes for a long time. But it’s a possibility.
Could it be anything to do with needing more RAM? Slows down when it hits the limit?
I’ve moved the temp folder to my D drive and since posting this it has increased from 38gb to 39.6gb, so it must be doing something right? I’ll leave it going over night and report back tomorrow.
The bar hasn’t moved though. I’ve marked it’s position with another app window!
thanks
more ram will probably help. but you can still get these issue with 128gb. i think it just doesn’t happen as quickly.
but think the real cause is generally one part of the mesh not being aligned properly.
so if you do a different reconstruction area that dosen’t include that part. then it should be OK.
or you go back and try improving the alignment.
Hi L P Whaites
And where it stops ? have you done some screenshot where it stop working or ?
So it stops in the same place everytime what made me think ram. Would more RAM extend the amount it can calculate before becoming super slow?
I need to look into reconstruction areas, don’t know what they are to be honest (in terms of utilising them). The Normal seemed fine which is why I didn’t question the alignment. The cameras look right to the eye and the pointcloud looks good to my eye atleast.
https://gyazo.com/b1825deca9244583b136221c21e279 - where it is currently
https://gyazo.com/bca12732486670416a5bbab410a41e3b - the blue is how much it has moved in about 10 hours. Normal detail was under 2 hours. It’s currently going to take days to complete. The others I have done were about 90 - 150 photos and took probably 3 hours-ish max high detail
thanks
if you haven’t been using reconstruction areas. then i would go back redo it with a slightly smaller area.
I’ve had plenty of these issue in the past. and a lot of the time it would work on normal. but high failed.
generally have done things in the past where i split the reconstruction area in many different ones. (though you should create overlapping mesh and cut it back later. otherwise you will get grid lines through your mesh).
after doing that it came back that just one of those areas was causing issues.
doing that will also help needing less ram. though overall it will probably take quite a bit longer.
450 images should not be a problem at all. It could take a while though.
Did you open the console window? There you can see more details, for example all the individual model parts and when they are finished.
What resolution of the images are we talking about?
What I can see from your screenshot is that your RAM is completely utilized and therefore RC has to swap data in and out from the terribly slow hard disk or ssd.
That will increase the time it takes for a project to finish to at least ten times the projected time if not more.
So you either are using very large source images or you are processing a large area and forgot to set a reconstruction area around the space or object you are interested in.
Also, RC is not supposed to run out of ram, so there is something else wrong…
Yeah you’re right, I found this https://support.capturingreality.com/hc/en-us/articles/115001483771
Which says it shouldn’t use virtual memory, ever. It feels like that is what it is doing so could be a bug. That is a much bigger model (5000 photos) but he does have more RAM. Possibly the same issue?
I’m going to try reconstruction areas now as it’s still barely moved and is going to take weeks.
I think I will get more RAM too. Once I do I will re-run this same project and report back with the results. That way we should know.
Interesting, I thought that Virtual memory was used, and even measure it’s peak usage for the benchmark - Maybe I’ll leave that in for initial testing or take it out.
The cache is effectively a more “non volitile” form of virtual memory in a way I guess.
From my testing, I have found that capturing reality can use
Cache folder
Project Temp Folder (different to cache folder)
RAM
on occasion Pagefile.
Yeah I’ve seen other posts refer to virtual memory as though it is used. I’m a little confused but hopefully more tests will help us figure out what is going on.
I’m currently rendering a region about 1/4 of the whole thing. It is already past the mark of the last one. It has slowed down but I’m assuming that is due to the depth pass counted for some of the beginning of the progress bar. Currently at 1hour 35min and about 2/3 there
I’m also looking for a script to cut up the reconstruction region automatically. I haven’t found one yet. As far as I can tell to render it out section by section you would need to export the text file, edit it and re-import it in. Otherwise it wouldnt be ‘pixel perfect’ - as you’d be moving the region by eye, I assume.
I’ve seen plenty of virtual memory used.
if its not meant to be using it. then its finding bugs in rc.
its worth keeping it in.
also keep in mind that last part of reconstruction is single threaded and so can take a little while so worth letting it run.
as for splitting up scenes you have 2 options. one is to write a script to generate all your regions. or the other way is you can type in the sizes/location in the info panel in rc. which isn’t a bad option if your only doing a few.
if you do split it. make sure you add overlap. you’ll have to clean that all up later. but seem lines in mesh can be a big issue if you don’t
The issue with VM has been discussed several times and the official answer was always that it does (or should) not use it. But it DOES happen more often than it should. So it would be really helpful if we could track down the reason why it is doing it sometimes. It has helped me once to reduce the size of the individual parts that RC splits up the mesh in.
L P, you could open the console window which displays much more detailed info on the processes that are going on…