I would like to try making 3d models from photos. I have attempted to use meshroom, but it requires an Nvidia card. Bought one, plugged it in and trashed my computer
Before I buy a new computer with the proper video card, has anyone here used meshroom to successfully make a 3d model from multiple photos and then cut from that model. If so what was your experience?
Depending on how complicated the model is, it could be extremely difficult starting from scratch⌠but itâs possible! I am not the expert youâre looking for, but I paid someone via Reddit For Hire to model a chess knight for me from pictures. So I sent him a bunch of pictures like this and he made this in Blender.
Ha⌠so after writing that, I was about to add something like âBlender (which I am assuming is similar to meshroom)â and then googled that to confirm and realized: Nope! Youâre not talking 3d modeling at all!
Now that I know meshroom is about photogrammetry, I remembered this video from Frank Howarth, and turns out he was using meshroom. Maybe youâve already seen it already? I have colleagues who also use this technique at work (Iâm an engineer working in product development) with success. Itâs sensitive to specularity (shininess) and is aided by random small dots. The colleague using this will paint a base coat of white matte primer, and then spray or flick with a brush to get random contrasting dots to enable the software to track these points as the item rotates.
I will say that the downstream process with the output is probably pretty similar for the knight I obtained from Blender and what youâd get from meshroom. I use Fusion to import from Blender, which exports a mesh (vs. a parametric model or solid body like iges). Iâve found stl and obj work fine, but I havenât had success editing much. Fusion wonât recognize it like a solid body, so you have to work with surface modeling techniques (of which I have ~0) or boolean operations.
In any case, Iâve successfully cut this knight just fine even though itâs a mesh body. I also found I could still extrude supports for 2-sided flipping and even though the additional CAD model is technically âfloatingâ vs. the model (none are connected), itâs worked great.
Someone is sending me a casting of a different kind of knight and I was going to try straight CAD on it⌠but maybe I can try this technique as well to see how well it works. Would that be of interest for you? Iâd just need to verify theyâre ok with me painting this casting with dots.
Ok, lots there but maybe something is helpful with respect to your inquiry. Given that I might be able to try this, is your main question something like âdoes this program actually work?â Or is there a different aspect youâre looking to validate? And what kind of object(s) did you have in mind? Perhaps my knight isnât a good use-case due to small size⌠I could try something else if your question is just to find out how well this process actually works. Let me know! Itâs a cool technique and Iâve been looking for an excuse to try it
Thanks. I started out way too complicated - I was trying to do a motorcycle. That didnât work but I have been learning as I go. I moved on to a model of a car, but I think that was too shiny. I am going to try to do something smaller and dull just to get stuff working.
I guess what I am looking for is the workflow. It looks like meshroom to blender and then hopefully I will get a model I can import into vcarve.
I donât think that meshroom will output anything useful or small enough to import. I think you have to use blender to remove the extraneous image and to reduce the number of meshes.
I assume mesh refers to each facet of the 3d model. Itâs like I need a mesh for idiots book LOL.
Ha! Wow indeed, a motorcycle is bold and adventurous for sure
Gotcha re. extra stuff attached. Indeed, Blender should be good for that. You could also look at meshlab. Iâve used it very minimally (just to convex hull things for robotics collision models), so I canât speak to which might be easier to use/learn for cleaning up the mesh.
My chess knight is ~15mb as an stl exported from Blender, for reference. Blender has a decimate modifier that can reduce the total number of faces, which helps a lot with file size. Itâs interactive, so as you apply a % reduction (e.g. 0.75 to be 75% reduced), you can look at some key features and observe if they noticeably decrease in quality.
Mesh, I think, refers to the whole thing. I think itâs just called a mesh as itâs a big network of triangles forming the outer surface/topography. You have nodes/points, edges that connect three of them at a time, and then a face/facet, which is the planar surface bounded by each of three edges.
If I take this on, Iâll try and remember to take some pictures/screenshots and describe what ultimately worked. Good luck!
You could also look into Zephyr 3Dflow, the free version allows for 50 photos so good for smaller objects, small price to upgrade to the Lite version for up to 500 photos. Nvidia card is recommended but not necessarily needed. For objects like a car with flat features and uniform color, try sprinkling some powder on it to add âtextureâ that photos.
Thanks - never thought of sprinkling powder on it. I did upgrade to an Nvidia card in order to run meshroom. I will probably look into 3Dflow after I try meshroom a couple more time
In case it helps, there are iPhone/iPad apps that use the FaceID or LIDAR sensors to create 3D models. Itâs not as flexible or accurate as photogrammetry but it is dead easy. I use Heges specifically.
On photogrammetry, the New York Times put together a detailed guide on how to do it with mobile devices.
I tried photogrammetry taking pictures of a portion of a tree trunk, gave lots of texture for the photo stitching with the bark and curve of the tree trunk. I was able to export the file and import it in Vectric software without much difficulty, simulated the cut but did actually cut it. In any case, you are more likely to get better result not starting with something like a vehicle on your first go.