I take a 360 video of a object, convert the video into frames and feed it to Reality Scan. When I align the images it shows up all the cameras surrounding the object. The AI masking masks the image very well, it would be great if it could also detect the main object from the mask and set up the reconstruction region on the object based on it.
Is this something that could be done currently or would it implemented soon?
Hello @Invigoretail
You can use AI masking tool before aligning. Then the region will be set only around the reconstructed object.
Thank you for replying!!!
I tried that, what ends up happening is that it discards a whole lot of images sometimes even less than 10% remain when doing alignment and therefore not all details are captured of the object as they are when doing it manually.
For example, if I have 300 images of an object which completely takes 360 view. When doing alignment it takes all of them into consideration and gives maybe 290 cams but when I do masking first and then align it gives me 100 cams but the cams don’t surround the object, most of the time a quadrant of cams are discarded. It also shows poor alignment quality message when doing masking first.
I also tried to give it 1000s of images of the object but it does poorly as compared to 300 but aligning first and manual reconstruction with the 1000 makes the model better.
Even if there were ways to have it reconsider the cams so that it doesn’t discard a chunk of them would be great.
OK, what kind of object have you captured? As only part of the images is aligned, then it could be inappropriate object for photogrammetry or the overlap between the images is low. Also, using video will provide worse model as using images.
So, it looks like the background of the images is helping in the alignment process, not the object. In that case you can also use the legacy masking workflow (which can end the same as AI masking mode): https://www.youtube.com/watch?v=xxsdbzBwWLE
What kind of object have you captured?
Also to your original question. there is a new reconstruction region setting which you can use after alignment: Set Region by Point Density
This should create the region around the most dense part of your sparse point cloud.
I am trying capture a 3D model of beverages as of now. I take a 360 video and convert it into frames. There is not much motion blur in the videos for it discard a chunk of images.
As per your suggestion, I tried setting the reconstruction region using point cloud but it does not fully zone in on the object and has point cloud of the background as well.
The software is able to create a good model when I do it manually so I think the object is good for photogrammetry. It is only when I do masking first and then aligning that it fails to pick up the points.
I saw the video that you shared, in that he sets up the reconstruction region on his own and then he exports that mask to be used later.
I can share the file if that helps.
The ideal case would be that it use the camera from the alignment for getting the poses of the object and then uses the masks to get rid of the background from the alignment so that the reconstruction region is perfectly set to the object. I think this would automate the process of creating a 3D model even further.
OK, so this is not ideal capture.
The scanned object covers only small part of the image and it looks to be quite featureless (big white parts).
You can choose in which step the mask should be used. Just select the images in 1Ds view and then change Enable masks for… to a wanted step:
So, you can disable them for alignment and enable them for meshing.
> This is with mask and alignment, you can see that the number of cams have gone down and the 360 deg is not fully complete
Okay, I try to do this but could not get it right will try again
(post deleted by author)
That works perfectly, thank you!!!
What would be the command in CLI for me to set the masks to not be used in alignment?
You will need to select the wanted images first by -selectImage command and then use -editInputSelection command with a key inpMaskOpts and wanted value.
