AB3D writes:
If you are interested in photogrammetry, Meshroom is a great option. However, you need CUDA for it, so it won't work on computers without an NVIDIA graphics card. I found a way around this using Google's free Colaboratory (colab.research.google.com), which gives you access to cloud computers. This has the advantage of being able to run on any computer, and it won't slow down your computer while Meshroom is running! I hope you find this tutorial helpful.
3 Comments
Hi AB3D,
This is really great for me as I don't have a GPU in my PC - thus an opportunity to do 3D scanning.
So I tried this great COLAB tool that you shared ..
However I could get a full 3D model for 1 case, but not for another case each with about 50 photos.
Can you advise why this is so? Any parameters I must set to get a full 3D model ? (see my sailor boy attempt only half a 3D model just like the sample Buddha, but the japanese model case was full 3D model. Also quality not too good ??
The pictures I had taken ares very clear .. but the model came out not very clear under interior studio lighting.
Any clue why this is so?
Here's the link of the 2 attempts of "toy model" using your approach ..
https://youtu.be/pxrw4Mot6jQ
Hi,
I’m glad you are finding this video helpful! From watching your video, it looks like your second attempt has reflective paint on it. Reflective objects are hard for tools like Meshroom to calculate, because they can’t understand what is being reflected and what is part of the actual model. This is my best guess to what is happening. For your current setup my advice is to try softer lighting, because I can see the lights reflecting on different parts of the model. If you need more advice, there is a subreddit dedicated to photogrammetry, so I’m sure you could get some answers there. Best of luck!
Hi,
thank you so much for this skript and the possibility to run it in an NVIDIA + cuda enabled cloud.
It works just as suggested in your video. However, I am trying to get only the point cloud (SfM dense ?) from this as my computer is not able to model that. I need the cloud information for a different skript and that's only able to handle a point cloud .ply file not one that includes the faces.
Can you point me in the right direction how I can run meshroom in colab and saving the full data folder and output structure that I would get from running meshroom on my computer - instead of only saving the .mtl and .obj?
Maybe that's too much to ask but I wanted to give it a shot.
Thanks!
Friederike