For those asking "Surface Reconstruction Poisson" is now under Filters > Remeshing, Simplification and Reconstruction > Screened Poisson Surface Reconstruction Octree Depth is now Reconstruction Depth Also "Quadratic Edge Collapse Decimation" is now called "Simplification: Quadratic Edge Collapse Decimation" in the same menu
Another metod of getting a picture sequence to work with, is this: Shoot a video (or find a suitable video clip elsewhere) where you circle the motive. Load the video into Photoshop. Export the video render BUT as a picture sequence. save all the pictures in a folder. In Explorer you size explorer so that you have thumbnails in sequence but with e.g. 10 columns of pictures in the viewpane. Box select the left column from top to bottom and copy those pictures to a different folder. You now have every 10 frame saved. That will decrease the number of files to work with. You can repeat as necessary.
Great video Gleb, thanks. One quick tip that will slightly speed up reconstruction is that VisualSFM has to be told to use CUDA for its calculations, it doesn't do it by default. In one of the menus (I don't have it nearby to check the exact names, sorry) there is the option "use CUDA" and a few parameters to play with. It should make a difference for you.
Very cool. I did basically the same tutorial a couple years ago but I like your addition of decimation. I didn't know MeshLab has that function at the time that I did mine.
Wow, so cool that you're doing all of this with freeware. I do similar work, but the software costs thousands. If I was younger with no extra money, I'd be watching this video multiple times and getting all the software listed. Too cool!!!
Please Help! All works fine with the example fotos. But everytime I take pictures on my own in VSFM I end up with only 3 Cams (of originally 92 ...). What am I doing wrong or is there some special trick while taking the pics? (Only tried tiny stuff like a PS Controler e.g. and ofcourse I did it 360° from various angles so they all should be able to get matched ...)
in VSFM all good after Matching doubles but something wrong after Reconstruction step. It allways matches like 3 or 4 fotos together before it starts over again from blanc ... :'(
photogrammetry is so much fun, its even more amazing that you could do it directly on your mobile phone with software like scann3d. though i think this workflow will give better results in the final render!
Happy finally you decide to try out free/open source solution vs 123 catch. Great tutorial like always. You can do the Poisson surface reconstruction also with CloudCompare , for some complicated model you can get better result than MeshLab. Personally when i do Poisson surface reconstruction i set to 12 8 1 1 and when apply Quadric Edge collapse i check preserve Topology Just a small terms mistake, if you do a camera movement closer to the subject you DON'T zoom but you do a dolly mouvement (Dolly mouvement : motion towards or motion from.), when you zoom you change focal length and the consequence is you change also the perspective ( is already a sort of special effect), if you do a dolly mouvement , or in short a dolly, is more natural movement for human eyes.
This is a great way to make very cool models of real-world objects like rocks and other natural stuff, Gleb! That is for a great introduction. Must. Sleep. Now.
Thanks Gleb, Really Exceptional Video. Your delivery is concise, informative and you speak better English than most native RU-vidrs. Thanks for taking it into Blender. I'm looking forward to your other videos.
Dang Gleb, you beat me to it! I was planning on making a tutorial with a similar method next week. Anyway, great tutorial! Definitely better than I could do.
Actually zero dollars means low cost at the acquisition. But the technology that we need especially for eg: computer, processor, cpu, gpu, and also the main idea of close range photogrammetry to create excellent 3D models are EXPENSIVE. It's so true that i felt it.. #vivascience #vivaphotogrammetry
Freakishly awesome Gleb. I work in the 3D field with 3D scanners and printers. I found this tutorial quite usefull as I have a few large scanning jobs I need to do this month.
I'm starting by saying that this tutorial is a life saver! BTW i have now the MeshLab 2020.06 version and I need some explanation about some commands that are changed over time.
The second step(reconstruct sparse) doesn't place points in 3d. Camera points placed correctly. But no model just empty space like object shadow surrounded by color points in one plane
I have a computer where the Sparce reconstruction only shows half of the images. But in another computer the sparce runs good. The first one has a GTX 650, the other is a nvidia 920A
Really Awesome Tutorial, so much humor and fun way to present information! Especially the idea of creating a paper millenium Falcon :D so great!! Maybe Nowadays (2018) I would use Meshroom as a one-click tool that extracts more points and converts everything to a model in one step.. Btw. one question: I struggle with photogrammetric Indoor Room"Scanning". How can I scan a whole room with this technique? A video for roomscanning would really help me so much oO
Hey Gleb , any particular things to keep in mind while taking photographs , optimal resolution of photos , what type of lights to use , is flash can be used (i guess not , it will ruin natural lights) , any other things ? Thanks for tutorial , will share!
Note that VisualSFM is only free for non-commercial use! There's an MIT licensed photogrammetry program at regard3d.org, though so far I can't seem to make it wrap all the way around objects, it insists on reconstructing from only one side.
Thanks for this tutorial!!!!!!. I've tried a few times to get this workflow right, with limited success. Maybe with all the details you have put into this video I'll finally get a usable import for blender
Great video! This helped me a lot, but I still have a problem. When I compute missing matches it says 0 pairs to compute match. I don't know why this happened.
Thanks for this tutorial. I didn't work quite well for me but this might have something to with my camera. I don't know. I see you alse made a tutorial with Photoscan but as it is older, i believe i wil use this method. Btw congratulations for your little shrimp.
I opened project in meshlab and it's empty, model gone , that got me confused initially, it seam that when its reconstructing its also saving which is untypical vis a vis photoscan
Hello Gleb, congratulations for the tutorial, really crazy. I have a problem. When the amount of interpolation in MeshLab, the matter at hand, upside down, and this does not allow me to create the mesh properly. Then beyond that I do not charge the intersection of the three axes, but moved to the right, how do I fix? thank you
Thank you for this, I've tried it a couple of times, but I think there's always a step that I've forgotten, and since I've upgraded the PC I used I will have to go back to the install step :(
What are the advantages of this workflow over the one with just Photoscan and Blender? Do the scans have better topology / less holes ? Is it faster? or is it just the fact of everything being free opposed to photoscan? Keep up the great work!
Thanks for the reply! After watching your videos, I started to focus on photogrammetry and blender, and now, after 6 months, I work professionaly doing projects also using real-world objects. The future is here!!
Great tutorial, I'm getting some nice results. Does the software use the accelerometer reading (is it stored on the picture file??) or its 'just' math? If so, we could keep the camera fixed and turn the object
hey this is a nice video and probably the first one who showed me how to correctly install the program. i still have problems with the result. the pictures from the third layer where you go as near to the ground as possible dont show up in the 3d modell. can u tell me why this happens to me everytime?
It's not showing all my camera locations in the first part of creating the sparse points, it only creates a few in a line at the front of the scene so it's not working properly. [EDIT]: Turned out it was my photos, I needed to take more clear shots and much more of them so the program can do it's thing properly.
Hey there, Gleb. At around 4:00, you copy and paste some files. You're using a PC for this. The files look totally different on my mac. What do I copy, if anything, and where do I paste them using a mac and not a PC? Thx.
When i try to use visualSFM it makes multiple "models" and matches the images that should only be for one object into multiple other groups, how do I fix this?
so lets say hypothetically I went overkill and had like 5k fairly low res photos. it is taking over 24 hours to just analyze and match the images. Should I even continue to the next step or just reshoot? I was trying to map an area of a building
Thank you for this, did wonder how to go about photoscanning. If taking photos of objects with sky etc in the background would you mask out the unwanted sky in each photo before starting the conversion?
