How to Turn/Scan Objects Into 3D Models With Just a Camera 

Alan's Art Log can you do this with drawings of a character that you drew on paper so then you have a hand drawn character?

Alan's Art Log so then you have a hand drawn 3D character* I mean.

@@AN-rc7xw I just realized this comment had no answer, but I had answered you. I'm not sure why my reply isn't here and it's nested elsewhere. Anyways the answer is no. I mean you could scan the paper but it would only show a flat plain with an image.

@Vanille Shake Sorry, not sure what you're asking. You tried this method and it didn't work? Or? I need more details to help.

Thanks. I am planning to do a follow up to this video (you can see my most recent videos where about making the new sculpt I'll be scanning), concentrating more on how to capture textures, and then after that I planned to show how you can transfer those to retopologized/shrinkwrapped meshes. It might be a while but I'm planning it. I'm always testing out different software (both free and not) and although I'd give a limb to be able to afford some 3d programs, even if I could I would try whenever possible to use the free stuff just because I know the feel. Also I've invested an insane amount of hours into customizing Blender's interface to make sense (planning a video on that too). and I refuse to stop using it now, lol.

Yes, I mentioned this briefly in the video. If you move the original pictures you imported, you will get an error. They become tied to the project in a sense. So you will need those as well as the project files. To fix problem, pick a folder to store your original photos, then open the .r3d project file with a text editor and find and replace all the old paths with a new one. (back it up first just in case). So if the pictures shared a parent folder say "C:\Users\Pictures\Scan Pics" do a search/replace for "C:\Users\Pictures\" and change to the path of the new folder they're in. Careful with the trailing slash, either exclude or include from both the find and replace.

I took about 70 photos initially, then added another 10 later because I was getting some artifacts near the top/bottom.

Sorry for the late reply. I don't know why that was. It works for me on my new computer now, and just in case it was the better hardware, I did test it on the laptop again (after a clean windows install) and everything worked. No idea why it didn't work before. It's also much better than the global reconstruction engine.

See the bottom of this page www.regard3d.org/index.php/documentation/details/picture-set it describes how to edit the camera database and also links to the latest version. You can go directly there to see if your camera is listed github.com/openMVG/CameraSensorSizeDatabase/blob/master/sensor_database.csv If it's not you will have to google to see if you can find the sensor width and manually add it.

I have, you can find the info at the end of the description. Note though this was on a laptop. For upcoming videos this will change. I have since switched to a more powerful desktop and everything is much faster for me and I have had no crashes or severe freezes.

Not much, you still have to do the surface reconstruction to get the model whether you want colors or not. What you can do is set the colorization method to vertex colors instead of textures. That should be faster and less memory intensive. Also you can ignore any of the tips I give in my follow up video regarding capturing the colors/textures correctly. So you don't have to have as complicated a setup, no need for a white background, or even lighting to avoid shadows (so long as the pictures aren't shaky/blurry and you don't have areas in complete darkness).

Yes, all things equal, the higher the resolution the more detailed the scan, the lower, the lower. As for trying this, you should be able to batch resize the sets (along one edge, either height or width, to keep proportions), there's many programs that can, and so long as they keep the exif data (so the program can match up my camera model to the focal length database) they "should" at least load. Also look up the kermit photogrammetry sets / examples, those are low res if I remember correctly and you can see the difference in the level of detail. Also note that the program might not be able to recognize your camera's focal length (check to see if the openMVG database has it), otherwise you will need to find it and add it manually to the database.

Although the proportions seem accurate (at least to what I can measure ~1mm), the object has to be scaled to the correct size manually using a real measurement along one of it's axis as reference, this seems to be around a factor of 10 (e.g. in Blender I have to scale x10, in Photoshop it has to be in cm and the numbers multiplied by 10) . And note if you later export from one program and import to another the scale might be off again since different programs handle sizes slightly different.

I found FSSR created more artifacts near the surface, especially where it couldn't "see" like beneath the chin and the ears. And even on the best settings I could find, those artifacts had holes in them, which makes them harder to fix. If I remember correctly Poisson is better for making airtight meshes, which is why it didn't have any holes, except for the mess at the bottom. What works best though probably has a lot to do with the pictures. I have not found a way yet to do an ideal scan (I don't have enough diffused lights), but then most people don't, which is why I went ahead and made the video. When I do I'l like to do a video on how that is better/different. But what I do know is that many of the sample photos available and probably used to test the program with are in ideal conditions and they're not full objects. One is like a partial view of a toy kermit, another one is an archway if I remember correctly. It's the types of pictures an archaeologist might take on sight because many of these programs were intended to be used for research like that. So maybe for those sorts of pictures FSSR would work better. I know for example that Structure from Motion won't fail with the example pictures, but do with mine.

Um, do you mean my images, or if you can in general 3d print the models it creates?

Yes, that was my intention, although I haven't gotten around to building my printer yet. It requires some cleaning up of the scan (so it doesn't have holes), but it's possible.

Yes, at the time I made this meshroom did not exist and this was the most user friendly. Even after it came out, meshroom did not have any documentation for a while. I was going to test it when the docs came out, but never got around to it and have been busy with other stuff.

I didn't know that. I'll pin your comment. Thank you. I'd heard of the paint splattering but not the dusting. I will give it a try. Is it a special type of powder or do you think something like baby powder will work?

baby powder will work and many errors you are getting are becouse of your computer graphics configuration, I think.

The comments in this section are extremely helpful.

Top Websites - www DesireFx .me 3D Free Model & Art

Do they all load and have the focal length set? If they do, then yes, there might be a bit too few depending on the object.

It depends on what you want to do with it. For complex objects that require complex deformations (e.g. elbows/arms) you will need to retopologize the object manually. See my reply to Robert Thurman in the comments for a more detailed explanation of what you might be able to get away with.

Unfortunately it's the only one I know where the author made it a point that it tried to avoid patent issues completely. See here: www.regard3d.org/index.php/documentation/license. From what I understand it's the matching algorithm (usually SIFT) that is the main problem (regard3D uses A-KAZE instead). openMVG + openMVS come close but openMVG uses SIFT by default. I believe there's an option to switch to something else but it's a command line program which makes it inconvenient. There's MVE which has a use interface but again, it uses SIFT in some part, not sure if you can avoid it. I didn't try either because Regard3D actually uses some code from all these (the parts without patent issues) so I didn't even bother with those other ones.

It depends on the step, some I know are resource intensive (Computing Matches, Densification) and I mentioned that, but you also don't need to set those too high to get good results and my computer was more than enough to get a decent scan which is why I listed it's details in the description for reference. I could max out it's RAM to 16GB but it's not necessary and there would probably be certain options where setting it past a certain level of detail it would still crash it. Of all the reasons I'm planning to update my computer, basically this is not one of them. There's other steps though that I don't really understand why they fail, like Incremental Structure from Motion. I said it crashes, that was a slip of the tongue, it just throws an error after a few seconds, but seems to work fine with the sample image sets on the program website. It's just my images it doesn't seem to want to play nice with. And the texture problem is indeed a bug.

Commercial solutions are always going to be better at this (maybe as you've noticed not UI-wise but in the actual processing). Apart from the things I mentioned in the video, you'll have to tinker yourself with the different settings, see if you can find anything that helps. Also you might want to look and see if meshlab has any cleanup tools that can help (I have a tablet so I honestly just prefer hand cleaning to meshlab's UI). Or maybe, I don't know if it lets you, I've never used ReCap, but can't you import the point cloud or mesh and clean it there? What are you trying to scan exactly? I wanted to scan more stuff but I haven't been able to play around with the software since I made this video because my fan is on the verge of breaking so I try not to do graphics intensive stuff. When that's fixed and I can scan more stuff if there's anything else I find helps, I'll do a follow up video or blog post or something. Also the developer seems to be active again and hopefully by then the texture bug will have been fixed and there might be other improvements.

the Sensor width doesnt show but the focal legnth does

No, Regard 3D handles everything up to the surface reconstruction. It does not handle any cleanup steps and does not allow any manipulation of the mesh, as you can see at the end, this has to be done elsewhere. Also size can vary between programs depending on how they import. Some let you choose. For me the size was off by a scale of around 10, and in programs that prompted for the size upon import/open, setting the scale to cm and setting the factor to 10 or multiplying the given numbers by 10 worked, but I have no idea if that's universal or if the scale is accurate. I'd have to find some calipers to check properly.

I have tested the size with calipers and scaled by 10 is just every so slightly off, it's a bit smaller, but otherwise, the proportions seems really accurate, at least to within what I could measure with some cheap plastic calipers (~1mm), they were correct.

There are special scanning sprays for this (I think I mentioned in the follow up video. I do not have access to them so can't say how well they work for this sort of photogrammetry, but I did test watercolors in the follow up video and those seemed to work (they're opaque and washable). Gouache also probably works well. They're a bit hard to get onto smooth objects though.

Well I could ask you a bunch of questions, see if we might find a problem: How many images did you take? How close together where they? Are you getting matches everywhere like with mine where the dots are scattered pretty evenly and there's so many lines between pairs that you can barely see anything, or are they more scattered? What options did you pick. etc. etc. BUT if you want it'll probably be quicker if you just send me an email with your images (zip them or put them in a folder in google drive and share them with me or something) and I'll take a look. If it looks like they should work, when I have some time, I'll try to run them myself. Email is alansartlog@gmail.com

Hello Alana thanks for the reply! I ended up using Remake and it worked easily. Although it is not super smooth. Form your experience is there an advantage in the mesh quality using the Regard3D comparing to Remake?

I've not tried Remake, I needed a free solution, but from what I've researched in general the scanning itself is pretty accurate for most programs, some might have a bit more noise than others. The differences are that paid programs will have more constant maintenance which means less bugs and more features, specially mesh cleanup tools. And I would guess they would also be faster because the developers have taken the time to optimize them.

@Alans Art Log Can we get your images to play around with the program for starters? It be really nice, because I understand it took along time to set up the light, background and having the sculpture spin consistently a certain degrees per image. Would also be nice if you can tell us how you know how much to spin the sculpture per image. But if you provided the images shown here we can get a feel how the program works much better. So that be super helpful!

The photos are my own of a sculpt I made of a character of mine. If it were of some random object I would have shared them, but I do not want to give these away. Also they're not that great because of the lighting issue I discussed. If I find some way to get even lighting while photographing around an object, I will do another video and make sure next time it's of something I can share. If you need sample photos you can find links to some sample sets on the regard3d website and I'm sure there are more floating around if you google "photogrammetry sample set pictures" or something like that. As for the angle, I think I did about 20 photos per spin. So about 20 degrees between each but I just kind of eyed the rotation, moved a cm or so each time. The more you take the better, but you also don't want to overdo it unless you have the processing power to handle too many photos.

The photos are my own of a sculpt I made of a character of mine. If it were of some random object I would have shared them, but I do not want to give these away. Also they're not that great because of the lighting issue I discussed. If I find some way to get even lighting while photographing around an object, I will do another video and make sure next time it's of something I can share. If you need sample photos you can find links to some sample sets on the regard3d website and I'm sure there are more floating around if you google "photogrammetry sample set pictures" or something like that. The ones linked on the program site though are more likely to work (if you'll be testing different settings) because the program was probably tested against them.

Well your comment is quite old and maybe in the meanwhile you already found a cool way to lit an object, but here is my suggestion (first project of the video ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-V5uycGosYq4.html). I didn't tried it with photogrammetry but it seems to be useful IMHO, hope you like it!

I tried that, but I unfortunately do not have a sensor that is in the database

Please see the follow up video. It has links to a different but better dataset.

Is the image info loading or is it blank?

It can work but there's two problems, you often lose a lot of resolution (to the point of things not working) and you can lose camera info on the pictures (which has to be added back manually => not fun) depending on how they're extracted if the info even exists in the video file in the first place.

It also gives he the error "Please wait until all image infos have been listed

The "please wait until all image info...." warning is normal, if all the info hasn't been loaded, wait a few seconds, you should see the metadata being added in the columns. The triangulation will fail error is weird though. Do all your images have the focal length AND the sensor width set? Does the camera model show up in the metadata and what is it? (to check it's in the sensor with database).

Actually my images don't have the sensor width set. It just says N/A. My camera model shows up, and it is Nikon D3400.

I'm only getting like 27/77 of my cameras and I don't know why...

I'm assuming you figured out how to set the sensor width to the correct value? It can be a lot of things. You might not match all the camera 100%, but usually for 77 I would get at least 50 matched. What settings did you do to compute the matches? Are you seeing a lot of matches (image is filled with dots/lines) or only a few? Are you scanning something shiny? Does it have enough texture?

It needs actual photos since it needs information about the image (focal length, etc).

@@alansartlog any idea how to upload custom screenshots into it to create a 3d mesh?

@@RCsFinest Um, did you not see my reply? You can't. "It needs actual photos since it needs information about the image (focal length, etc)."

@@alansartlog I did read that I thought there was some sort of plugin that could import custom screenshots

Is the button missing completely or is it grayed out? And did the triangulation and densification succeed?

Alan's Art Log it is not greyed out there is no button at all I even tried re-downloading the software

Is there also no cancel button? Do you have a high DPI set by any chance (i.e. your OS's UI scale is set to something other than 100%)? Though it's strange even if you did that you didn't already run into the issue in the first step like I did. Still you can try hitting tab until the selection reaches the end then disappears, then hit enter. That should hit the okay button if it exists but is hidden. You can also turn on text-to-speech (Narrator in Windows), it should narrate the button when you tab to it. If none of that helps, if you can send me a screenshot (alansartlog@gmail.com) I might be better able to help. You can link to one as well, BUT links tend to get caught in the youtube spam filter and then I never get notified.

I have not tried this but you should be able to get it working by following the instructions at the bottom of this page. Basically you have to search for the sensor width online then add the info manually, just make sure you spell everything right (I'm assuming it's listing the model and make just not the sensor width) and it should work. www.regard3d.org/index.php/documentation/details/picture-set

Hi Alan, thank you for reply I tried Sony Sony Cybershot DSC HX60V 7.76 as my model Sony DSC-HX60V For whatever reason saving the file and restarting Regard3D still says sensor NA, I think I may need to ask developer as it is odd it won't accept standard format. Thank you for help

Did you make sure to separate that with semi-colons and that it was exactly the camera model the program was seeing? If so, if you don't mind sending me a picture, just one will do, (my email is alansartlog@gmail.com) when I have some free time I can have a look. I dabble in programming here and there. If there does seem to be a bug I can submit a github issue for you.

Sony;Sony;Cybershot;DSC;HX60V;7.76;;; Sony;Sony;Cybershot;DSC;HX60;7.76;;; Sony;Sony;Cybershot;DSC;J10;6.16;;; mine is top entry- I'll grab camera from work and send you a photo- I tried most likely options similar to the list of sony's and restarted program without success thanks for help

I think the format is "Maker SEMICOLOR Model SEMICOLON Sensor Width SEMICOLON" so Sony;Sony Cybershot DSC HX60V;7.76; not semi-colons at every space like your examples. Hopefully that works, otherwise I await your email.

Was your model not in the repo mentioned in the link (this one: github.com/openMVG/CameraSensorSizeDatabase)? If not you should just be able to google "samsung YOURMODEL camera sensor width" if it doesn't work with the phone's name, use it's model code/number. There's also this wikipedia article, might have your model listed: en.wikipedia.org/wiki/Samsung_CMOS

Yes I've tried VisualSFM and a few similar programs. It works pretty much the same except you don't have a built in dense cloud creator (you have to download CMVS/PMVS seperately and link to it) and you don't have a built in surface generator, which means you need to use something else like meshlab to actually create the mesh. Also there's no branching or anything. The UI in general is a bit more confusing. And the license is not really open source (there's an article that tends to come up when you search for it that incorrectly implies it is, but just search for the license, it's not free for commercial use). I choose to use Regard3D because this video was aimed more at people who might have never heard of photogrammetry, and it had a good UI and the developer took extra care to only use libraries that wouldn't have copyright or patent issues, see www.regard3d.org/index.php/documentation/license. You should be able to get good results with most of the photogrammetry programs out there (some might require more tweaking than others). Many of them use the same libraries under the hood (e.g. PMVS/CMVS). There's a guy who wrote a few blog posts testing a bunch of different programs (although not Regard3D) and he made a cool little chart (pfalkingham.wordpress.com/2016/09/14/trying-all-the-free-photogrammetry/) where you can see all the different program paths you might take, and you can see VisualSFM is the most complicated while Regard 3D can do all the steps (though of course mesh cleanup is not included in the last one, just texturing). Hope that helps :)

thanks man.

Man you are great.

Yes. You can read the license here if you have any doubts: www.regard3d.org/index.php/documentation/license. Good luck!

Would I be able to do a whole body shot or just face ?

You can capture whatever you like. The only limits are your lighting/environment and your computer. You'll have to consider the following problems (almost regardless of what you try to capture): - Where are you going to take the pictures? It needs to be well lit and evenly lit if possible. Also the blanker the background the less noise you're going to have to deal with in post-processing. If you can't find somewhere suitable your next best bet is a setup like mine (one wall, hang a sheet if it's not blank, then stand on a turntable) but you might run into the lighting issues I did. - How are you going to take the pictures? Can your subject stay still? Usually in studios this is done with dozens of cameras (or a laser based system) at once so it's quick. But I'm assuming you just have one camera so you'll need someone to take the photos and you're going to have to not move with arms outstretched for a long time. If you want to capture your clothes it'll be even harder. - How much detail do you really need? You can get pretty good results with low polys and good textures. I'm assuming not a lot because this is for a game and will be the equivalent of a sculpt (needs to be retopoed, rigged, etc). But if the answer is a lot you'll likely need to do your head separately. That is, try to take the body photos all from the same distance where as much as possible of you is in frame, process those, then do the head closeup separately. So in light of all these obstacles have you considered just doing the head, retopoing that, then stitching it to a free model? With just a front and side picture of yourself you can then adjust the body model with inflate/move tools, and you'll have saved yourself the trouble of having to retopo it. Otherwise if you do decide to go ahead and try to capture your full body I would suggest taking just one 360 round of shots (10-15 shots), see how well it's capturing with your setup (is there noise? do you need to be better lit? is it capturing too much of the background), adjust as necessary, retest until you're good, then take the final set of photos. That should cut down on overall processing time and frustration. Hope that helps. I'd love to see how it turns out :)

do you mean its saying your focal length is missing? what step exactly do you get stuck at?

Same here. When you upload the images, even though focal length shows up it say triangulation will fail. Which it does and nothing shows up. Ive used my dslr and my iPhone and same response

Yes, it was because I'd misread it. If you liked the video btw, be sure to check the followup one: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-7pGJ8rWiecw.html

he also made Poisson distribution, I just read a paper about it.

As you can see in the followup video (ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-7pGJ8rWiecw.html), unless you want to capture the background, it's a bad idea. If there's any color in the background, the object gets tinted, and even if there weren't, usually you don't want to capture the background as it just means more cleanup.

I'm using a Canon DSLR, not a phone. It doesn't matter what your focal length is though, as long as it detects it you're fine, it should still work. Typically there's no way to change a phone's camera focal length since they do not have a physical zoom lens like a professional camera, but seeing that phone has several cameras with different focal lengths I have no idea how that works. Sorry. If it's writing the wrong focal length to the file (which seems highly unlikely) there are probably programs that can edit EXIF tags (where the focal length is stored). Just search something like EXIF tag editor.

@@alansartlog But my triangulation always fails because of that

What error are you getting exactly.

@@alansartlog As soon as i place the pictures it says "the focal lenght is known for too few pictures" "triangulation will fail"

@@alansartlog Anyways i proceed but as soon as i want to triamgulate it doesnt let me and i get stuck at that part

Mmm, no never tried the B&W. Interesting idea though. Will give it a try when I have some time. About the size scaling, I have thought of that, but haven't tried it yet because it would mean being careful to capture two objects, and a cube would need to be made out of something with enough texture to capture. One easier idea I had recently was to print a square grid over which to lay the object. That way it doesn't get in the way. Have not tested it yet though. But the precision is something I'm interested in, specially comparing it to a high quality laser scanner (which I might get access to in a year or two through my university) so in the future that is something I would like to test in depth.

Laser scanners are what I would like to use too but for hobby work they are so expensive. What I want is a scanner that will give me the 3D surface map without any texturing (labels, colours,etc). I was working on this technology in the past and we were getting close with stereographic cameras in B&W but there’s a big jump from looking at objects and interpreting them as 3D solids!!

I see the scanning of objects as 2 processes. #1] object scanning and solid modelling from point clouds to polygons and #2) texture mapping.

Well, technically, if you factor in the cost of a good camera, you could probably build a small 3D scanner for half the price, probably less. In my case I already had my DSLR though and I'm assuming you too already had your cameras. Don't know how a cheap scanner would compare to photogrammetry though. And I haven't really investigated in depth why some scanners cost so much, how they work exactly, etc. But I am interested in building my own although that's a project for the far future. Currently I'm focusing on saving up for a 3D printer. Once I have the printer, it would make it easy to print the parts for a laser scanner, I could even temporarily cut costs by borrowing parts from the printer (motor, threaded rods, etc). Regarding the textures, why does it matter that it not give you textures? You can easily delete/remove the extra information like I did.

I have an Ender3 3D printer. About $180. Worth the price IMO.

Well this will help you scan it but you would still have to clean it enough to make it workable, retopologize it, and rig it. For easier/faster retopology in Blender you might want to check out retopoflow. Also look into the remesh and the shrinkwrap modifiers for quickly doing things like arms (also cleanup). If your character is simple enough or has few deformation points or it's movement are restricted (think a penguin) these can go a long way (this video: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-rUzJefYNQlI.html is a good example of how those work and goes over some other options). If your character doesn't require deformation (a robot) you might actually be able to get away with 90% using those modifiers. For rigging, Blender includes some base rigs (humanoid, animals, etc), and there's probably even more available online (search Blendswap). There's no magical way around this if that's what you're asking. The scanning is a replacement only for the sculpting step. Also ideally you should have sculpted your character in the rigging pose otherwise I wouldn't recommend even trying, especially if you're unfamiliar with Blender/3D software. Oh and I see you mentioned it's a cartoon. If you're going to want to render it in a stylized fashion (NPR) in Blender I would suggest you check you can get the look you want first (also the regular Blender renderer that's usually used for stylized work is being depreciated in 2.8 in favor of the EEVEE renderer which I still think is missing some features for NPR).

it is

I just checked and the program is using an older version of the camera database. Try downloading the current one here github.com/openMVG/CameraSensorSizeDatabase, it contains the iphone 6. Or copy the entry for your model and add it to the camera database manually. You can find detailed instructions about how to do that at the bottom of this page www.regard3d.org/index.php/documentation/details/picture-set. After that it should work.

Thank you for your help. I am still not able to make it work. I installed it on a Mac computer and a windows 10 computer. But on both I was unable to find the camera database at all located anywhere on my hard drives. Thanks for trying

Does the folder just not exist at all or is it just missing the camera database file?

To me it looks like it does not exist at all. I used the searched the whole C drive for the database file. No search results showed up at all. I looked for the folder and was unable to find it. It may be hidden from me but I have my settings to review all plus file extensions. Thanks for your help. Best of luck to you. Lee Noring

I'm going to use windows as an example because I'm more familiar with it, but I'm assuming you're having trouble finding the AppData folder. A search will not find the file because AppData is a hidden folder and nothing in it is indexed in search. There's ways to get it to show in explorer, but the easiest way to just get there is to either 1) hit Windows + R then paste %LOCALAPPDATA%/Regard3D and hit enter. Or 2) paste that into the explorer path bar and hit enter. You should open right into the correct folder or if it's truly missing it will give you an error.

I'm not sure what you're looking for exactly? Research regarding the algorithms the programs uses? or?

yes i wanted to know algo program uses because i wanted to build 3d model from images taken from a camera with my own algo

The algorithms/libraries the program uses are in the step names themselves (AKAZE, MVE, Poisson, etc). I think Regard3D's site has a detailed list of libraries it uses somewhere, you can start there.

I just used two desk lamps with soft-ish neutral led lights. Nothing fancy. In a new video I will be publishing soon there will be pictures of the actual setup which I've improved by making fabric covers to soften the lights further, but still nothing fancy. The four lights was just an example of what might be a more ideal setup (I don't have enough lights for it) but I'm no longer sure this is the case, not the number of lights, but the camera moving around the object (see new video when it comes out). The more lights and the softer they are, usually the better. Actually I should say the more evenly lit the object, the better, because a harsh light could make things worse. Or if the lights are only overhead, like a well lit room, because they're overhead, you'll get hard to remove shadows in the textures (if you care about them). In the new video I did some experiments regarding the background. Plain backgrounds are best, otherwise you get more noise. And white backgrounds are best if you care about the textures (colored backgrounds can bleed into the textures). But if you can't get around it, you can still get decent results. It's not the biggest problem.

Ok Thankyou But i want to go with this idea in my project but am confused about some points. . Is it compulory to use model of only one color? and i want to ask you if u r working further on this field will u be improving the model if yes then through which new techniques u r using? If i ask u about any special research paper regarding this then what will be ur suggestion? I hope u will be understanding my point..

Regarding single color models, no, it's not necessary. Regard3D is capable of capturing objects with textures and everything (depending on the surface). I go more in depth into the best surface in the upcoming video. The results I got were a bit weird (contradicted some of what I thought in this video), it's hard to explain in text. For now I'll just say unless your object is glass or smooth plastic, just try to scan it as is (no paint splashes or anything), see if that works first. The new video should be up in a week or two max. It will cover which are the best surfaces, 90% of the settings with visual comparisons, and compare different light/camera setups. I'm not a researcher or anything btw, just an artist curious about using photogrammetry to scan stuff.

@@alansartlog i hope this will work out with me more

I'm working on a followup to this video that might help (for example, since making this video, the incremental engine now works for me and works far better than the global, there's also a few other things I've learned that might help). In the meantime though, if you want to send me an email (alansartlog@gmail.com) with what the problem is exactly and if you can either a link me to your photos on drive or something or send me a screenshot of some of your pictures, I can try to help, tell you if the problem is your photos or the program settings.

Alan's Art Log Thanks, will definitely take you up on that when I have some time!

Yes and no. This is actually something that is used for scanning things professionally, especially actors, look up something like "professional photogrammetry setup", and companies like ten24 use setups like these to create models, but as you can see they usually use way more than 30 stationary cameras (100+ usually). With 30 you might be able to capture one side of something, like a face, maybe the front of a body (depends on the resolution). One work around might be to create a rotating rig with the 30 cameras in 2-3 rows, and then rotate them around the person and you can scan them quite fast, though not instantly like with these stationary setups which can also be used to capture things that move like animals. You might also want to look into actual laser scanners but I'm not too familiar with how they work and there must be some downsides if usually photogrammetry is used for scanning actors as far as I know.

Yes, as I mentioned, I /thought/ this would be better BUT I actually tried this (as best as I could) in a new video, check it out here, it goes way more in depth ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-7pGJ8rWiecw.html (there's a Table of Contents in the description) and the results surprised me. Turns out it's much more important to have good even lighting with few shadows than just about anything else.

?

I didn't understand what you said, @samljer, but speaking on boolean operation performed in the last part of video: well, why did you use a plane to cut the model instead of subtracting a cube without the hassle to later remove the plane's points and the point behind the cut?

Thanks. I have already been made aware. I completely misread it the first time and it stuck in my head.

? It's not contradictory, maybe it confused you, but you do not need to save. All progress is automatically saved. So not sure why you think it's wrong...

wth, looks pretty clear to me