Combining components created with masks

Comments

9 comments

  • Avatar
    Ondrej Trhan CR

    Ring could be hard to do, as there is hard to find texture. How does it look?

    Did you try to use Merge components option?

    Is it possible to you to merge the datasets partially, not in one step?

    You can also use some control point to merge these components, if you will find some good places to palce the control points.

     

    0
    Comment actions Permalink
  • Avatar
    Simutech21

    It's a ring with a round stone but both material are pretty smooth. The two separate models look okay, a dent here or there (nothing I wouldn't be able to fix afterwards), just the thin part that would be underneath your finger wasn't reconstructed properly in either model and not sure I can do anything about that.

    I just ran "Merge Components" (under "Alignment") but it added 8 more components.

    Sorry, what do you mean? I've got two image sets that show the ring from the top, which I used to create the first model, and two image sets that show the ring from the bottom, which I used for the second model.

    How many control points do I need? The smallest component only has 2 cameras (0), the largest has 140 (8). If 0's images don't share any points with 8's, can I "link" it to a different component through control points and then link that one to 8?

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    I thought, that you used all images in the project in one step. What I suggested is, that you will add the datasets separately in more steps (one step for one image set).

    It is hard to say, how many CPS you will need in this. But you need to have minimally three CPs in one component. These CPs should be the same also on another component and CPs should be placed on minimally three images from each component.

    In this case also could help using the chalk spray, which will create the texture for the ring. Then you will have two datasets, one for alignment and second one for texturing.

    0
    Comment actions Permalink
  • Avatar
    Simutech21

    I've got 3 projects:

    1. The top part (= 2 image sets from different positions, using a turntable)
    2. The bottom part  (= 2 image sets from different positions, using a turntable)
    3. Both top and bottom

    I created a normal model each for 1. and 2. and used those to create the masks, then combined them in 3., hoping that the masks would be enough to combine the top and bottom part, like in the tutorial I linked. They weren't.

    There are a couple of scratches to the left and the right of the stone, so I used those to create 4 control points and then added them more than half of the images (wherever they're visible properly). The problem I'm now left with is that the scratches are a bit blurry in a bunch of images that show the back of the ring because they were focused on a different spot, so I'm not sure how to handle those best. If I just guess, then I get a lot of red warning triangles for specific control points for specific images (probably because the positions are slightly off).

    I deleted the old 9 components and aligned the images again (with the control points), hoping for the best but now I've got 22 components. :/ One of them uses about 150 images and it looks like it combined the top and bottom a bit but it still doesn't look as good as it should. There's also another component with 70 cams but that's just the bottom part and I'm not sure why it wouldn't combine it with the other one.

    Unfortunately I can't retake the images, as it's not my ring and even if it was possible, I wouldn't be allowed to use any type of spray either.

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    As I wrote, this is quite hard dataset for any soft. On this type of images is quite hard to find any features and that is why you get so many components. Also, as you wrote, you have unfocused areas on some images, which also is not ideal.

    If there is no option for re capture the images, then you need to use more control points. is quite small number for such object.

    The mentioned spray is possible to wash out easily with a water.

    Is it possible to share your dataset to watch it what could be wrong here? 

    0
    Comment actions Permalink
  • Avatar
    Simutech21

    I added more control points (up to 11 now), mostly small scratches that are only clear-ish in around 10-15 images but that are visible in at least 2 or 3 images that also show the main 4 control points but I think I messed up:

    I managed to get it to use 260 images within a single component (the others were only used in around 15 small components), then tried to set a couple of more points in those unused images but now I'm down to 240 used images and the model looks like multiple different-sized models were put together. :( There are also a lot of red warning signs with values of 50px and more, which I don't understand because to me it looks like they are in the right spot in the images. Could it be that RC has problems putting together a point if it's set in the blurry part of an image and also in the sharp part of another image? Of course I didn't make a backup when it looked semi-good and now I can't go back.

    Sometimes RC suggests a new control point's copy's position, is there a way to reset certain points in certain images to that position?

    Sorry, I can't share everything, as it's multiple gigabytes. I can upload a couple of images/screenshots though, what would help the most? Something that shows control points in different images?

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    How many images did you captured together? Is it possible to somehow simplify and send me these simplified images?

    This tutorial could help you to follow the principles of placing control points: https://www.youtube.com/watch?v=S00_mLfbx6o&lc=Ugzj_RLQ3M5IzRGdNHZ4AaABAg

    The ring problem is, that it seem very similar from different sites. You need to be 100% sure about the placing of control points. About the red triangles, the CPs could be on right places, but when these images are wrongly aligned in the component, then you can have these big values. You need to achieve to have it smaller than 1 px (ideally).

    Yes, also blurry parts could be a problem, as the features detected here could have different positions and it is not so precise. You shouldn't use blurred images in photogrammetry (in these cases are objects captured with different focused parts and then is created an image just from focused parts).

    If you want to use suggested position, you can delete this measurement and then suggest it by RealityCapture again. 

    The best way are the images, but if it is not possible for you, you can send some prints screens of placed CPs and how the components look. 

    0
    Comment actions Permalink
  • Avatar
    Simutech21

    There are more than 350 images, plus the masks.

    That's the video I found too but sadly it doesn't mention too many of the problems you can come across or give any suggestions about how to fix them.

    About the red triangles, the CPs could be on right places, but when these images are wrongly aligned in the component, then you can have these big values.

    I see. How can I make sure that they're aligned correctly?

    I now started a completely new project and redid everything but slower:

    I loaded the first set (top), and picked 6 images, then used 3 of them for the same control points for the scratches on the left side, then 3 images for the right side + a point in the middle (instead of setting the CPs in most images). Then I loaded the next top set and assigned the same CPs. Rinse and repeat for the 2 bottom image sets. Surprisingly enough, Reality Capture then suggested more locations for the points and (apart from 3 or 4) they were all in the correct spots. The "Align Images" result now uses about 230 cams and it correctly put together the top and the bottom part but the inside of the ring has what looks like a lot of loose vertices that then mess up the whole model. It's especially bad in the "normal" model because the ring "melts" into the surrounded box.

    I uploaded an image with a couple of screenshots here.

    Regarding the blurry parts I was talking about: Look at the inner part of the ring in the very first image. It's out of focus but instead the "scratches" are in focus. A couple of images are the other way around: Inner part is focus and outer part isn't. There aren't any images with both parts in focus at the same time, so I can't even do what the guy did in the video because I wouldn't be able to link the two parts properly because of the blurriness.

    Regarding the other CPs I mentioned: In the second and third rl image there is a dent in the middle, just above the stone. It's not visible from the side though, that's why I didn't use it this time. Same thing with the inner part of the ring: There's also a dent and there are even a couple of scratches and an inscription but they're only visible (and in focus) in a couple of images, that's why I didn't use them this time.

    Two questions:

    1. How do I get rid of these flowing vertices?
    2. The brown thing you can see in the back of the first rl image is a wodden tetris piece. There are 4 that were also on the turntable when the photos were taken (they're in almost every image). I used the box in RC to limit the scene to the ring but is it possible to still use those wodden pieces to create more control points (at least only for the top and only for the bottom, as the pieces weren't flipped when the ring was)?

     

    0
    Comment actions Permalink
  • Avatar
    Ondrej Trhan CR

    Yes, the video was created for ideal case. As there are not these problems, them it is not needed to solve it in the video.

    Two main problems here: the ring is too shiny for photogrammetry to find enough features and you have unfocused images. The melted part of the model is caused by these unfocused areas, as it is hard to define exact position of point for this.

    The alignment depends on the way, how the object is captured. These errors could be reduced by using another images or using more control points with higher weight.

    The control points are too close together, you also need to have some on another part of the ring.

    About the unfocused images, you need to do something like there is described to achieve better results: https://www.ifolor.ch/en/inspire/photoshop-photo-stacking-for-continuously-sharp-images

    About your questions:

    1. You need to recapture the object. Or use just focused parts of your images to create a model. The flowing vertices are direct result of unfocused images.

    In solid mode you can use Select marginal triangles in Scene 3D tab/Tools/Mesh Model/Advanced and filter this selection. Also, you can filter bigger triangles by Select large triangles tool.

    2. Yes, you can use these pieces to align, but just for those images, where ring is not moved. So I think for just top images and bottom separately. You won't connect these two components using them (if they have still the same position)

     

    0
    Comment actions Permalink

Please sign in to leave a comment.