Obrigado :)

i do not get it working can you explain very precize how to do it? (sorry for bad english and bad skills)

Sounds like densification/meshing may not have worked. Do you have an Nvidia GPU? If so, it may just be failing on the reconstruction - does the sparse cloud look good in COLMAP?

What are you using as your workspace folder? It might be that you have spaces or other invalid characters in there.

Details on hardware and images available on my blog: pfalkingham.wordpress.com/2016/09/14/trying-all-the-free-photogrammetry/ Basically 4 core i7 (with hyperthreading), GTX 970, 16gb ram, and photos taken with a Sony Nex-6 (16mp).

@@PeterFalkingham that's pretty impresive with just the power of a 970

@@daton3630 yeah. I mean, the video's a few years old now and the 970 wasn't as long in the tooth then. I'm on a 2060 super now, and it's not much faster, I think the CPU is the bottleneck now.

You don't need to put anything in there, you can leave it blank. I don't use masks for images myself, so can't tell you much about it.

The Megadepth dataset maybe?

Nope. Bear in mind that the normal version of COLMAP requires an NVIDA card, because it relies on CUDA, so is incompatible with a Mac even in virtualized windows. You could try Colmap-CL, which replaces CUDA with open-CL: peterfalkingham.com/2021/02/15/photogrammetry-testing-colmap-cl-a-game-changer-for-those-without-an-nvidia-card/

@@PeterFalkingham Thanks: you're really on top of things!

Do you have a space in the path? I think you have to have a full path without any spaces. e.g. "C:\A Folder " won't work, but "C:\A_Folder" will.

@@PeterFalkingham I will try it. Thank you very much.

Sounds like you might have a space in the folder path?

@@PeterFalkinghamThanks for your advice. Thanks to your tip , i was able to know that my GPU was not enough to proceed it.

Afraid not. Not a lot to go on, but it sounds like it wasn't able to match any photos.

@@PeterFalkingham I just used a lot more images. Worked .

I don't have this dataset anymore, but you can get my other standard dataset from here: peterfalkingham.com/2020/07/10/free-and-commercial-photogrammetry-software-review-2020/

No, that's not normal... You should be able to get as far as the sparse reconstruction without an Nvidia GPU (I think). Your 4gb of RAM is probably the biggest issue. How many photos are you using?

@@PeterFalkingham I am using 75 photos, the quality is set to medium. Thanks for your quick reply!

@@kevegonczi2835 I'd start debugging with something like 10 photos (make sure they are all taken very close so they can match). That'll tell you if your system is limiting you.

Colmap should just find it - you don't need CUDA libraries installed, just a CUDA compatible GPU with recent-ish drivers.

I don't have this dataset anymore I'm afraid, I captured it specially for a sketchfab blog post (sketchfab.com/blogs/community/tutorial-using-free-photogrammetry-software/) but didn't keep it. If you just want a dataset that is known to work, I have a set of 53 photos available from my blog - head here: peterfalkingham.com/2020/07/10/free-and-commercial-photogrammetry-software-review-2020/ and scroll to 'with my standard dataset'

Then you don't have enough overlap between images. Have you tried on a dataset known to work?

COLMAP is quicker and easier to use, but only takes you so far (it won't produce a textured mesh). I review both (and other photogrammetry software) on my website: peterfalkingham.com/2020/07/10/free-and-commercial-photogrammetry-software-review-2020/

Hard to say. Are you getting a decent reconstruction in the main view?

@@PeterFalkingham That happens to me too. I can't find the file I need for the Meshlab. I see the reconstruction image well.

@@HesTiaFuLL Did you work this out?

Did you work this out?

Do you have an Nvidia GPU? Sounds like it's not getting past the sparse reconstruction stage.

@@PeterFalkingham I don't have one. I took the version that does not require you to have CUDA. What's the point of having a version that doesn't require Nvidia/Cuda if it's useless?

@@batvanio Are you using the non-CUDA version from the COLMAP site or COLMAP-CL? The former provides enough to get a simple model, but the latter, available here: github.com/openphotogrammetry/colmap-cl Can take you all the way through to final model. More info: peterfalkingham.com/2021/02/15/photogrammetry-testing-colmap-cl-a-game-changer-for-those-without-an-nvidia-card/ If you are using COLMAP-CL, then not sure why it would stop at the sparse reconstruction.

@@PeterFalkingham The first. I didn't even know he had what he mentioned. I'm testing it now. Thank you for the information.

This produces a mesh through photogrammetry then dense meshing. I have a written sort-of tutorial for Instant-NGP, Nvidia's free NERF software: peterfalkingham.com/2022/04/26/neural-radiance-fields-nerf-and-instant-ngp-future-of-photogrammetry/ - that uses COLMAP in the first stages, then produces a NERF. It's not easy though. For something easier, I suggest signing up to LumaAI's work: lumalabs.ai/ It's in beta and cloud-based, but waiting lists aren't too long.

@@PeterFalkingham thanks a lot ain't luma lab an ios app ? i'm on android

@Sauveur its Web based, so works on android or desktop, just upload a video.

@@PeterFalkingham wow thanks

You need to do file->Export model as, then choose whatever you want to use, ply, obj etc.

Thanks for replying me..I want to get depth of that images in the form of png/ jpg.so plz guide me

@@anusuiyatiwari1800 Oh I see. Yeah, the depth maps are in a binary format, not sure how to convert them. You might be able to open them in photoshop, but otherwise not sure, sorry. I'll take a look when I get chance

@@anusuiyatiwari1800 If you open the dense reconstruction window, and make sure the workspace folder is set, you can view the individual depth maps and save as png. It looks like you can only do one at a time though....

@@PeterFalkingham Okk sir thank you I will try and then I will touch with you.

Mobile phone is completely fine, though I tend to use a DSLR for projects where detail matters (whether it makes much difference is debatable).

Not that I'm aware of, no.

@@PeterFalkingham Thanks for your help. Much appreciated.

I think TIFFs or PNGs work too. I know other software like Meshroom and Metashape can take raw files, but not sure COLMAP can.

This was on my old desktop, an i7-4790K with either a GTX 970 or RTX 2070 (can't remember), and 16GB RAM

Sounds like it failed to match images. If COLMAP doesn't present a point cloud in the main window after, then reconstruction failed at some stage.

Nope! Photogrammetry uses the image itself to reconstruct 3d positions. You just need lots of overlap between images. See my blog for more info.