Hacker News Reader

Comment by mysterydip | Hacker News Reader

Can anyone suggest a hobbyist/entry-level 3D scanner? I have a 3D printer that I would love to replicate or modify some real-world objects on.

View on HN

They all suck. I've tried so many from cellphones to $5k devices. Until you get into $20k systems, don't expect a workable scan that doesn't required many hours of CAD work for cleanup.

Photogrammetry will be your best bet. Also not cheap. https://www.autodesk.com/solutions/photogrammetry-software

I would really suggest learning Fusion 360 for mechanical or blender for organic modeling and build from scratch. It's a great toolset to know and can be very rewarding. It will teach your core skills that you can apply all the time with your 3d printer. If you have never done any 3d work before, start with tinkercad.com until you feel comfortable.

Tangentially related but I've had a 3D printer for years but only just recently learned Fusion 360; it has been a real game changer in terms of the kind of projects I can tackle. Can't recommend it enough.

Shapr3D was the game changer for me - although you need an iPad pro for that one

What did you use to learn F360? I’ve been meaning to get started for a minute. But when I have the time I tend to forget to fire up 360.

I used lynda.com myself. In many cities in the USA you can get free access to the site through your local library. I found it easier then searching for a video here and there on youtube.

I’ll look into it. I have pluralsight through work. But I’ll check the library for Lynda.com access.

Start with making rings. It sounds silly, but they’re small so they print fast so you’ll get to learn fast. Plus, people will enjoy them.

Containers is a great next step from rings. They’ll encourage learning about fit, coordinate systems, and support structures.

There's a two or three parter on youtube where some main Autodesk dude makes a widget. It's something like 90 minutes all-in, and that was enough to give me the traction I needed to start experimenting. From there it's re-watch bits as needed, and googling around---mostly ending up at the Fusion 360 help forums.

I think Solidworks is a better bet for numerous reasons:

1) The interface seems to be simpler, especially when making things a little bit more complex

2) Greater amount of CAD models already available in this format

3) Greater amount of resources and tutorials

I believe Fusion360 is free if you are making less than $100k/yr on it. And the skills are transferable to Solidworks if you continue with CAD.

It may be free, but it's still proprietary

You think? I could never figure out how to do anything in SolidWorks, whereas I could just pick up Fusion and create. Fusion's interface seemed much more intuitive to me, and I had zero experience with CAD before. Plus, it's free.

Nowadays I use OnShape because it's decent and the only thing that works on Linux. I do miss some of the more advanced features, though (e.g. simulation, or variables).

SolidWorks is $5000/seat, right? At least, for the non-Student version?

And the Student version is only available to people with a valid current student ID, and saves files in a format that is intentionally incompatible with the normal version?

not a solution, but their entrepreneurial pack will set you up with a year long seat on just about whatever software they offer for free.

Student files are not incompatible, but as soon as a student licensed copy touches a part or assembly, it becomes "tainted" (such that it says "student edition" in the corner or something). I worked at a university-based company where we had a bunch of student licenses and a couple pro licenses, and we had to be very careful about separating the files and making sure that the student edition never touched our professional files.

I'd disagree with 2 and 3, at least at the hobbyist level. Fusion can import STEP files, which at least in my experience are far more common that SolidWorks files. And there's a huge number of Fusion tutorials and videos online, not to mention that Autodesk has a large selection of howto videos on their website.

Organic modeling in Solidworks seems to be extremely difficult. That being said, the Power Surfacing plugin has a good reputation:

https://www.npowersoftware.com/NewPowerSurfacingOverview.htm...

Skilled operators can likely work near-magic anyway, given sufficient time. ;)

Why don't depth cameras and SLAM techniques work?

Generally, the depth cameras don't provide enough precision in the data to be able to manage anything usable. If you want to scan an object to recreate something close it'll generally work ok but if you wanted to scan the object to build around it, and have things like up accurately then it's not good enough.

They do work, just not well enough to be useful for scanning. I have a handheld infrared SLAM scanner, and when it encounters moderately reflective or dark surfaces, it interprets them as voids because they absorb or scatter the IR. Anything translucent or refractive is similarly a nonstarter. The surfaces it does scan are not very accurate (covered in slight bumps from inaccuracies). This means it can't scan most objects, and the ones it can scan still require a ton of cleanup.

well it looks like its a multiview structured light and another type of camera.

So I suspect it is a SLAM system (how else can it make up an accurate mesh like that) but with a calibrated and well integrated depth sensor on there as well.

Seconded for Fusion360. Lots of great guides online and an intuitive UI. I started using it when i got a CNC router and eventually transitioned to 3D printing. I did not have a modeling background and felt comfortable making things after about a month

Totally agree. The only "cheap" 3d scanners that work even somewhat well are the ones designed to scan small objects via a rotating table and laser/diode array. I've had decent results with a friend's Einscan SP ($2,500) and a ~3 inch model, but awful results with every handheld I've tried under 20k, even at CES this year.

I’ve used an Einscan-SE/SP. How do the 20k+ scanners reduce the part-to-cad workflow? Is the software that’s included do a 10x better job of cleaning up the scan?

At the hobbyist level currently the best results (in terms of cost/quality and time/quality and overall capability) can be achieved using photogrammetry in conjunction with processing in Meshroom and post processing in Meshlab or Blender or editor of your own selection. This works for really tiny stuff right up through drones scanning entire buildings, and the workflow is the same. That's not to say there isn't a knack to it or things you have to learn. But it's the right direction to explore right now.

See: https://www.youtube.com/watch?v=ye-C-OOFsX8 and https://www.youtube.com/watch?v=1D0EhSi-vvc for a quick intro.

There are alternative workflows and "low cost" commercial hardware and software products even for photogrammetry in the market. Nearly every technique I have tried or seen produces a much lower quality result than the above, costs a lot more, or has some other limitation (such as the size of the object you can scan) that radically limit its usefulness.

This is of course until you get to the high end scanners like the one linked here which use one or more high resolution LIDAR sensors in conjunction with cameras, lights, motion sensors, etc: alll selected and calibrated to give precision results. If you can stomach the price they are awesome.

But to some degree, commercial solutions regardless of cost will often promise their output will require "no editing or post processing" You should be aware that this is total baloney. You will absolutely have to devote a little time and effort into learning how to edit, clean, and simplify meshes regardless of what you intend to do with them.

All this being said, this advice only really applies to objects where 3d scanning has some merit. If you want to 3d print something like a replacement part for a broken appliance, often 3d scanning even with an expensive commercial scanner is not the best approach. Many times it's considerably faster to pull out a pair of calipers and simply model the object in some CAD software like Fusion360 or for real simple stuff, even TinkerCAD.

Meshroom is really good, and it's even better for large scale stuff (like landscape / architecture reconstruction from aerial pics).

Nothing works. I promise. I’ve even used these extremely high end scanners like the creaform ones and they suck.

Here’s how I get _good_ results on complex objects. Use agisoft photoscan, a cheap SLR and a stepper motor controlled turntable. You need to put all this in a photo tent and light it as flatly as possible. Then, you need to do surface prep. Spray your object with magnaflux spotchek. This is a white powder in acetone which will go down on the surface and make it temporarily matte. It wipes off when done.

When you’re done, process the images, process the mesh, and pull it into Rhino to repair it. If you’re modeling for a functional replacement, pull out your calipers and take measurements and use your 3D scan essentially to pick off profiles that you can loft to remodel the final object.

This shit is really hard.

What where you trying to scan that the creaform one couldn't handle? I've worked with them for metrology and they are really good.

A knob for a vintage oscilloscope. It’s entirely possible that myself and the guy that maintained it in the architecture dept of my school just weren’t well versed in its intricacies.

My guess would also be that it’s small-ish for this kind of scanner.

Photogrammetry apps are your best bet at the hobbyist level. Might be tough with some materials or shapes, but it isn’t going to set you back $35k.

I assume it’ll only be a matter of time before this tech makes it into everyone’s smartphones. Sensors and algorithms are cheap.

https://www.aniwaa.com/best-3d-scanning-apps-smartphones/

Check out https://github.com/alicevision/meshroom Video tutorial from Prusa: https://www.youtube.com/watch?v=1D0EhSi-vvc

For $400 you can get the Occipital Structure Core [1] with Skanect Pro scanning software bundled. I've used Structure Cores for other purposes, and they seem pretty good. Not sure what your accuracy will be though.

1. https://store.structure.io/buy/structure-core

I had some minor luck with capturing 3D objects in Meshroom, then using Blender to refine/simplify the model, before taking it into Meshmixer to prepare for actual 3D printing.

I don't know how phone cameras hold up, but I was able to achieve decent results with a DSLR and some daylight lighting.

Perhaps check out https://www.open-electronics.org/an-open-source-3d-scanner-m...

As others have said, if you want to replace an existing part, it's probably actually better to redesign it. I use OpenSCAD to do simple 3D designs in code. It's easy to get started if you're a developer.

As for scanning, I've used Trnio on iPhone for artistic scans. e.g. I scanned my dog and printed a tiny replica. My nephew thought it was a pile of melted yogurt, but I knew what it was. :-)

If you have access to an iPhoneX and some patience, Scandy Pro is worth a try: https://apps.apple.com/us/app/scandy-pro/id1388028223?ls=1

Your cellphone. Iphone has the best 3d scanner apps, android has some good ones.

They use photogrammatry and do a fine job.

Got this in my emails recently: https://www.instructables.com/id/DIY-Budget-3D-Scanner-V3/

I think for entry level there are lots of apps for Android/iPhone that are passable for small objects but time-consuming.

Check out Structure Sensor (structure.io) if you haven't yet. Full disclosure I work for the company that makes it :)

The Galaxy Note 10+ has one, but I don't know how good it is.

The Kinect is okay as a 3D scanner.

Reply on news.ycombinator.com