I use RealityCapture. In preview mode, the render finishes without any error, the result is bad, but normal it’s previewx mode, but in Normal and High Detail modes, I get the error “computation of one or more part failed”, with 6 parts that have failed. It’s strange, because it’s always the main part of the model that is affected by the error. By the way, I don’t have this error when I want to use another set of pictures, for another model
Hi TheWonderPie, for the missing part one of the recommendations is to change reconstruction region (increase its size and rotate it a bit around vertical axis).
Other recommendations are to set Minimal distance between two vertices (Reconstruction settings - Mesh calculation) to a value; e.g. 0.0001 or higher. Also raise the number of Maximal vertex count per part (Reconstruction settings - Advanced) to a higher value; e.g. 10M or 20M.
Also, it seems, that you used video to create images. As the person could move during capturing, the results won’t be ideal in this case.
Thanks for your answer. Unfortunately, doing what you told me to do (Minimal distance between two vertices : 0.0001 ; Maximal vertex count per part : 20000000 ; Rotated and increased the size of the reconstruction region) rendered a completely blank model, so it’s even worse.
I tried to reset the settings, but let the reconstruction region bigger and rotated, and it game me a slightly better result than the original render, and when setting the Max vertex count per part at 3000000, I have almost a good result, but I do not know that this setting actually does, but you’re telling me to increase it instead of decreasing it, am I doing the right thing ?
Is it possible to share your data?
Thank you, I will also need image data.
I’m sorry, but unfortunately I have deleted the frames by mistake, but I have still the original video (it can take some time for Google to process the video)
Hi, I tried to download the video, but for some reason it is not possible. Is it still uploading?
Hello, yes as I said Google can take a long time to process the video, and it does what it wants, so here is the drive link, which is processed in priority :
Hi, this is my result with pre defined settings:
It is not great, but better than yours. What were your settings, when you imported the video? Also, I changed the reconstruction region to not model the walls behind the person.
Indeed it’s clearly better. I have set the seconds between each frame for extracting them to 0.3. By the way, I even tried to do a clean reinstall, but the same problem occurs.
And by the way #2, I noticed that it’s always the part of the model where there are the most details (the main part of a model) that are cut out in the end. ( https://imgur.com/a/WYYplU1 )
Can you try to set the reconstruction region just around of the scanned object and compute it?
Thank you ! It worked ! But what if I wanted to process a bigger model ?
Then you need to capture also those parts of the model. In this case, you was scanning the person, not the area around him.
Yes but when I do that, the person will be “bugged” like with my original issue
On the same subject : On a RealityCapture sample dataset (the Handmade Christmas Decoration one), I did set the reconstruction region smaller, just so it fits the basket, it’s better than when it wasn’t set around it, but some areas are still “bugged”, it seems that the fix only works with the person
It depends, how it will be captured. Models from video are always worse than from images for example. Also, you need to follow some rules, how to capture objects, like it is described here: https://www.youtube.com/watch?v=9e_NLp_FaUk&t=202s.
In your case you just captured correctly the person, not the area around. Therefore the resulted model is needed to have the reconstruction region.
Also you can try to capture person and area separately and then merge these areas.
We always recommend to use reconstruction region, no matter what. Which dataset do you mean? Can you send me the link?
I understand that there are “rules” for capturing photos, but samples are taken by the devs of RealityCapture, so it’s quite impossible that the issue is the photos
I was/am talking about this that one dataset https://www.capturingreality.com/free-datasets#:~:text=Handmade%20Christmas%20Decoration
I understand that the person was the only “thing” to be really captured, but even then, if I enlarge the reconstruction region (while keeping the person inside it), it’s not normal that the person gets bugged
For this case indeed it would be a fix to merge areas, but it just moves the problem : There are still models that aren’t fixed like that, like the basket from the sample dataset I sent you
Here is what I have for this case( You can see that the reconstruction region has been limited to the main object, and even though it’s better than what it was before doing so, it’s not fixed):
Hi, this is not normal behavior. Which version of RelaityCapture do you have? Is there something written in console view after computation? Do you have enough space on the disc, where is the cache?
When you mentioned clean install, did you do it this way:
- reinstalling and updating Nvidia drivers,
- removing cache (navigate to you cache location and remove the cache folder)
- " make it like a clean install" (this will reset your settings to default in the process):
In order to do that:
- Shift + Double LeftMouseClick on RealityCapture icon
- From drop-down select make it like a clean install and press ok
- Then try again
I have the latest version that was on the website a little more than a week ago, I guess there weren’t any new updates since
Previously, I had just did the Double click on the icon, and then select Clean Install, but not the other ones
So I just tried to do them, including selecting Clean Install once again, but it gives me the exact same result
Here is what is written in the console Yes I have enough space on the disk (260Go on an SSD), and the cache is on that drive, as well as the photos
Hi, can you also show me the 1Ds view with the wanted component?
Are you using for this Start button? Is it the same, when you use Calculate model after alignment?
In console it says, that calculation was aborted. Did you abort the calculation?