Help with Geometric details

Comments

25 comments

  • Avatar
    Götz Echtenacher

    Hi Justin,

    I think you got most of your questions right. Focus and lack of features due to the material. Geometry is not responsible at all...

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Thanks for your response.  

    So, I guess my new question is...how exactly is best to make sure its in focus?   I mean, yeah, I get it, "focus the camera so its in focus"  but what I'm saying is with images this high of resolution, part of the shoe is ALWAYS going to be out of focus--what's the best approach for photogrammetry?. This was shot at f/13 on a 50mm, which most the literature I've read said this should be fine.  On my lens I could go up 1 more to f/16 but still, it's never going to be in focus from front to back or top to bottom on the angled shots.  Never mind the fact that a shoe is a much more narrow subject than most (say compared to a bag of chips) 

     

    Should I always focus to the front edge of the subject and leave the back edge out of focus? 

    Would lowering the image resolution actually help?  As I mentioned, working with images this high res, you really start to notice things being out of focus that you would never notice on a 24mp camera.  

    As for the translucent material, worst case scenario I can spray paint the shoe with gray primer if that ends up being the bigger thing causing the issue.

     

    I promise I'm not being lazy by just not testing all these scenarios, I'm just trying to get up to speed with the nuances of shooting for photogrammetry  which would take days of testing and banging my head against the wall.  Just need a couple nudges to get me going in the right direction so I can start learning myself

     

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Btw, I did do an updated test today with more photos from the same set still bumpy looking but better.  I think this is getting closer to passable, especially with an SSS material on the sole, you probably won't see the bumpiness.  

    .  I'll shoot a new set soon and try again later

    Updated Results

     

    ***Edit*** screw it, I bought a cheap nikon refurb lens that goes up to f/38. I'll shoot a new set with that tomorrow and see if the the improved focus range helps my scans.  

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Justin,

    not to worry, it comes across that you don't just walz in here and ask silly questions.

    The images are not bad at all.

    With extreme f-stops you will introduce more blurriness, most lenses are best at about 5.6 - 8. For the most common ones there are plenty of reviews for that.

    Resolution isn't everything in my opinion. If you have less, you just neet to take more images. And to avoid your focusing issue you can do exactly that. One circle a bit further out or just vary the focus plane every second shot.

    You really need to try out what works best for you. Manuals can never cover anything an imho photogrammetry is more of an art than a science or purely technical application.

    Oh, and the primer might work but only if it's sprinkled - but I think the focus will sort those areas out...

    0
    Comment actions Permalink
  • Avatar
    Lukas Schulz

    Provided you have the time and necessary willpower you could also do focus stacking.

    Meaning, you take multiple pictures from the same position with different focus points and let software (for example Helicon) calculate a single sharp picture from them.

    It does get very time expensive though because the photo taking process rises by a factor of up to 10.

    I did notice though that you get away with a lot less pictures because the pictures you DO have are a lot more detailed.

    Definitely worth a try for objects with a narrow focus plane and pretty much the only way for very small objects.

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    @Gotz-- yeah, I was considering the issue about extreme depth stop and overall blurriness last night as well.  But I'll test out and see how it goes.  

     

    @Lukas I was thinking about this last night as well!  I have done focus stacking before with some of my product photography, but yes, it would take much longer.  I can do a small test with that and see if it's worth the extra time.  Maybe less photos, thus saved computation time, could offset the time it takes to focus stack.  Another thought is...I believe the Nikon D850 has an automated focus stacking feature.  I might be upgrading my D800 to that camera soon, so if I could give that a look as well.  

     

    New updates to come today!

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Yes, testing for oneself is always the best!!!  :-)

    Although I have no personal experience with focus stacking I have heard concerns about distortion geometry. The focus works by adjusting lenses, so each focus step would have a different geometry. If you then combine those in one image there will be no proper solution for RCs algorythms.

    If you do the stacking in combination with a slight movement of the camera, RC will do it all for you and you'll have the added benefit of more angles. Just make sure the distance between 2 cameras doesn't get too small. Since RC can handle thousands of images, I think you needn't worry about that. And I bet the time difference isn't great, might be in RCs favor anyway... 

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Quick test photo with the new $77 lens 😂  shooting at f/28 now and getting MUCH better sharpness with the expanded DOF.  I also picked up a ring light to supplement my current lighting setup.  I should probably get a circle polarizer for this lens to reduce some of the reflections happening but for now, I'll test this setup out-hopefully have a new model tonight showing better results in that tan sole.

    Test Photo

     

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hey Justin,

    Glad to hear it!

    Interesting info and thanks for sharing!

    Curious to see for myself, but the link doesn't work for me (neither does the one before that, the first two are fine).

     

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Try that previous link now, I updated it.  

     

    Here's the latest model from last night.  I didn't calculate a model for the whole shoe, as I always get an error about 40 million poly limit if I do the whole shoe (is there a quick way to limit the poly count on the High Detail model?  RC's settings for this are really odd and being that it takes a few hours to make a model, even for a small section, it would take forever to figure this out on my own.)

    Sadly, I'm still getting a lot of texture showing up in the pattern on the sole.  This shoe scan was made with 400 images total.

     I may switch to a different shoe for testing--one with a fully opaque sole--so I can figure out if the translucency of the sole rubber is the culprit.  As you can see, the sole towards the front is scanning a bit smoother, which is where the rubber is less translucent so I'm led to believe that may be the culprit.  

    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant

    Don't worry about the 40M "error", this is only a display warning. There is no limit, I already worked with 2B+ models.

    Just simplify your model to what you feel as acceptable after the high details reconstruction.

     

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    @ Jonathan_Tanant, that's good to know!  So, basically, just calculate the entire shoe then simplify it to something below 40m polys and I will be able to see the model on screen in RC?   Out of curiosity (I can't test because i'm on the demo version) does a model over 40m polys export fine--it's just that over 40m won't display in RC?

    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant

    Yes, that is the idea.

    If you want to export, 40M is quite big. You will be able to open it but maybe it will be too heavy to process as is. It depends on what you need to do with the models.

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Good to know.   Yeah, I know 40M is way too big, was just curious.  Probably I'll be export models out to bring into z-brush, or some other program, and bake out displacement maps to put on a lower poly model.  As an aside,I have a feeling I'll need to actually learn z-brush finally, to smooth out the noise in these models instead of getting it sorted out in the scan--sigh.   

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Justin,

    the link works now, cheers!

    I think that with that kind of resolution this is as good as it is going to get. There is a limit, in my experience, to the geometry level of detail where it is still acceptable that lies abou factor 5-10 below the pixel resolution. That means small details need to be 10-20 pixels wide in order to be modeled, more so if it needs to be perfect.

    So in your case you are doing really well imho. The reason for the slight bumpiness in the sole is probably rather the smooth surface than the translucency. If you want a more perfect surface there you need to zoom in further. There might be irregularities like that on the fabric as well but you won't be able to pick them out because of the strong texture.

    Just a thought: the way you balanced the shoe seems a bit precarious. Since a shoe is also a soft object, I wouldn't be surprised if some of the artifacts you observe are actually small changes in the shoe's position.

    RC will still display the model even at billions of vertices because it filters them to an acceptable amount that you can define in the settings. Only the solid and sweet views won't work for big meshes.

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Awesome, thanks for the extra info.   And yeah, any noise in the uppers, i'm fine with as it is fabric anyway, so you'll probably never notice that once a material is on it.   I'll do some testing with more closeups and see how that turns out.  I've also ordered a can of temporary spray that i've seen others use for 3d scans, so I'll test that out once it arrives.  

    yeah, i'm trying to figure out the best way to hold the shoe to scan it, to get all sides.  In RC can I just sit it flat for a few scans, then flip it upside down and rescan, and reality capture will align them (either automatically or with control points)?  I've seen this done easily in ****.  

    For the pattern on the soles, would it help at all to use a marker to place some dots on the shoe? Say, a red dot on each peak and a green dot on each valley of that y shaped pattern?  So far it's getting the general shape of that pattern extremely well, but I wonder if doing that it would help RC be less confused in those areas and maybe get a smoother result? 

    Thanks for the tip about simplifying to be able to see the mesh!  Here's an update of the full shoe.

     

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Justin,

    very nice result imho!

    I still think that with the resolution it is as good as it is going to get.

    The spray paint might indeed help, but only in the flat, featureless surfaces and almost certainly not for the writing on the sole - there the resolution is too small in my experience.

    Flipping it works well as long as there is enough common surface between the sets at more or less 90° viewing angle (as in parallel to the sensor).

    You might also want to play around with the distortion models and see if that helps anything - with such miniscule detail it might make a difference.

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Thanks for all the help.  Going to have a much improved scan to show soon--very excited about it! 

    Quick question about capturing all sides (this is a different shoe for this question)  

    I have 4 rotations captured with a shoe mounted like this.  Is there a way to do rotations of the shoe mounted differently (so that the tripod is on the opposite side, (so the whole sole is visible) and RC will know to use the clean sole data?  Or is the best approach to create models of both versions and delete/combine later in my 3d program?

    It appears I RC doesn't support deleting point cloud data, which is the method to do this in ****.  Ideally I would like to be able to have RC create the full shoe instead of me having to combine in a 3d program, as in many situations my object won't have have hard break points that I can use to combine multiple sections.  

     

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    I think the only way to get this working is to mask each and every image before feedeing it into RC.

    Here is a method that works without much additional effort:

    https://support.capturingreality.com/hc/en-us/community/posts/360043121532/comments/360005149732#community_comment_360005149732  (scroll down to the second screenshot)

    The trick is to have different backgrouds for each take and not have anything obscure the scanned object.

    Of course, you need a feature-rich object for this to work, which, as you suspect, may be a problem with the sole.

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    you mind re-linking to that other thread so I can see their method (you accidentally linked back to this thread).  

    Getting it all in one scan isn't a big deal for this shoe, since it'll be easy enough to either manually combine separate scan models or model it manually.  But for future scans it could prove useful to figure out a bit better.  

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    well that's super interesting.  Curious if you know the explanation as to why having a totally different background works (although it looks like it works, it seems to go against everything I've learned about RC so far).  I can't wrap my head around how this wouldn't cause RC to just not match the images completely.  Is it because RC will discard this background info since there is no overlap and only keep the main subject since it is detecting overlap in the main subject?

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Ha, excellent questions!

    RC seems to be doing things a bit differently at times compared to other software.

    I think it's a bit of a numbers game -  the surface that's on most of the images wins.

    It works similar when you get a noisy surface (in extreme cases a recognizable doubling), which is most often caused by misaligned cameras. The effect is imho caused by the two misaligned sets of cameras battling for superiority, meaning in some areas one group has more images, in others the other group and the algorythms then cause the surface to jump back and forth between the two possibilities...

     

    0
    Comment actions Permalink
  • Avatar
    Justin Goode

    Interesting.  I'll give it a shot next scan I do!

    I was just studying this guy's work this morning to try and decipher how he was getting such amazing results.  Looking at the viewport image he posted of his shoe scan, it appears he might be using your suggested technique to great effect (it appears the majority of the shots are of a shoe sitting flat but you can see some with it laid on its side).  He only used 195 camera in this scan and got way better results than my last shoe scan test using 700 grrrr  Oh well, I'm learning more every time!

    https://index.artstation.com/artwork/Dx54L9 

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Oh, it's Vlad! Probably one of the best out there...   ;-)

    You're not too far off though, I think.

    0
    Comment actions Permalink

Please sign in to leave a comment.