5 ms·

That's not how it works. 3D scanning and scanning-like techniques serve a very different purpose than actually 3D modelling the thing, and are in no way a subst

by 542458 10y ago

That's not how it works. 3D scanning and scanning-like techniques serve a very different purpose than actually 3D modelling the thing, and are in no way a substitute. - You need considerable 3D domain knowledge to employ the scanning tool correctly in the first place, and to post-process it's results - The scanning tools don't work in many situations - have fun scanning something transparent, reflective, that won't sit still etc etc etc. The curtains in this scene, or the translucent lamps, and so on would all be very, very problematic - The scan gives a noisier result than doing it by hand (Look at the model quality in figure 1, last step - it's pretty bad - that would be really, really apparent in a higher-res moving scene). High-end laser or contact scanners are better, but very expensive. - Scan topography is currently really bad from every suite I've seen, and not really ideal for anything other 3D printing or using as a reference for building a proper 3D model.

AndrewKemendo 10y ago
3D scanning and scanning-like techniques serve a very different purpose than actually 3D modelling the thing, and are in no way a substitute That's too broad to be true and your examples below don't say anything about the fundamentals of 3D scanning. It also ignores causality - in that some things are built in 3D first and then fabricated (extruded/milled etc...) while others are fabricated then scanned. So the 3D data around each can't really be compared in the same sense. In fact photogrammetry and laser scanning is getting to the point of eclipsing hand modeling for pretty much anything that is not a result of a CAD model. See: Our photogrammetry patent
- TrevorJ 10y ago
  3D scanning is a a great tool, but in addition to manual modeling. The scanned data is a great place to start, but if you want to animate the item, or use it in any sort of realtime or near realtime application you will be replacing every single polygon using a mix of manual modeling work and other algorithmic tools. Raw scan data is a fantastic starting point but it's not an end to end solution.
  - AndrewKemendo 10y ago
    * or use it in any sort of realtime or near realtime application you will be replacing every single polygon using a mix of manual modeling work and other algorithmic tools* Our product is real-time and "other algorithmic tools" are the basis for our work - and we have had fantastic unprecedented results. Raw scan data is a fantastic starting point but it's not an end to end solution. That is true for current COTS tools, but what we built has proven that we can get an end-to-end solutions for our problem which is opaque, non-reflective objects larger than a soda can and smaller than a vehicle. I know everyone on HN is the smartest ever and nobody here can see other people as making real shit but there are innovations happening. In this case the wide scale need wasn't there for innovation for a long time and now there is.
    - TrevorJ 10y ago
      That sounds fantastic and I look forward to seeing it hit the market. On the other hand, like you said, the state of the art for off the shelf tools isn't there yet so I think the snarkiness is kind of off putting.
- 542458 10y ago
  You're sort of lost me - I too do work and research in this field, and I have never seen the question "Do we 3D model or 3D scan this?", treating them as completely interchangeable. Can you give an example of a domain or application where that is true? The point of my examples was that scanning is currently so limited as to make a pure scanning workflow unworkable in almost all situations, including this one. Scanning is incapable of doing what modelling can (good topo, clean results, accurate representations of translucent, reflective, etc surfaces). Likewise, modelling is incapable of doing what scanning can (quick approximate real-world captures). They're different tools that accomplish different things. > In fact photogrammetry and laser scanning is getting to the point of eclipsing hand modeling for pretty much anything that is not a result of a CAD model. Due to differing definitions of what is considered "CAD" I'm not sure what you mean by this. I can think of applications where 3D scans are used, but only either as a reference for hand-modelling, or only after being so heavily hand corrected that it's a really more of a hybrid (model/scan) process. Animation almost always uses hand-modelled or heavily corrected (by hand) scans. Manufacturing uses hand-modelled objects or traced-over scans. Gaming uses hand-modelled objects or traced-over scans. Where are scans eclipsing hand 3D modelling? They're being used but as a different part of a hybrid workflow.
  - AndrewKemendo 10y ago
    Well in our case we use models of furnishings. The space of furnishings is so large that you can't hand model everything manually in polynomial time (that's a joke). As a result we have clients that want 7 million objects in 3D. So we had to figure out how we were going to build a massive model repo quickly and at high enough resolution to be good enough for rendering. Scanning is the only way to do that so we built an entirely new way to do that combining the latest photogrammetry and depth scanning research. Going further, with already built things, that aren't previously built in CAD, like components and old buildings, in solidworks autocad etc... Making those models is much more efficient with structured light or photogrammetry systems. It should be noted that it's not just speed, it's hard to model to micron specificity, whereas we can with scanning. Animation is a little different but I think follows the same principles in the long run. The distinction between the methods in my mind is between digitizing something real versus making something new. I think in the former case scanning will win long term.
- pen2l 10y ago
  > In fact photogrammetry and laser scanning is getting to the point of eclipsing hand modeling for pretty much anything that is not a result of a CAD model. Woah, interesting. Please go on? (i.e. what would be your workflow for this - what applications/apps/hardware would you use for this? Also, for a thing with a lot of polygons, like a human, ok -- but would you do same for an object better suited for CAD programs? Like a monitor, let's say?)
  - AndrewKemendo 10y ago
    There are a variety of them. As I mentioned elsewhere the core distinction is in digitizing real things that did not come out of a manufacturing process that utilizes CAD. For us, as we are an AR company, building 3D content is a critical piece so we need to digitize real objects as quickly and accurately as possible. The trick for us is finding the perfect balance between size of the resulting model and quality. So our scanning process is very refined.

by 542458 10y ago

AndrewKemendo 10y ago

3D scanning and scanning-like techniques serve a very different purpose than actually 3D modelling the thing, and are in no way a substitute That's too broad to be true and your examples below don't say anything about the fundamentals of 3D scanning. It also ignores causality - in that some things are built in 3D first and then fabricated (extruded/milled etc...) while others are fabricated then scanned. So the 3D data around each can't really be compared in the same sense. In fact photogrammetry and laser scanning is getting to the point of eclipsing hand modeling for pretty much anything that is not a result of a CAD model. See: Our photogrammetry patent

TrevorJ 10y ago
3D scanning is a a great tool, but in addition to manual modeling. The scanned data is a great place to start, but if you want to animate the item, or use it in any sort of realtime or near realtime application you will be replacing every single polygon using a mix of manual modeling work and other algorithmic tools. Raw scan data is a fantastic starting point but it's not an end to end solution.
- AndrewKemendo 10y ago
  * or use it in any sort of realtime or near realtime application you will be replacing every single polygon using a mix of manual modeling work and other algorithmic tools* Our product is real-time and "other algorithmic tools" are the basis for our work - and we have had fantastic unprecedented results. Raw scan data is a fantastic starting point but it's not an end to end solution. That is true for current COTS tools, but what we built has proven that we can get an end-to-end solutions for our problem which is opaque, non-reflective objects larger than a soda can and smaller than a vehicle. I know everyone on HN is the smartest ever and nobody here can see other people as making real shit but there are innovations happening. In this case the wide scale need wasn't there for innovation for a long time and now there is.
  - TrevorJ 10y ago
    That sounds fantastic and I look forward to seeing it hit the market. On the other hand, like you said, the state of the art for off the shelf tools isn't there yet so I think the snarkiness is kind of off putting.
542458 10y ago
You're sort of lost me - I too do work and research in this field, and I have never seen the question "Do we 3D model or 3D scan this?", treating them as completely interchangeable. Can you give an example of a domain or application where that is true? The point of my examples was that scanning is currently so limited as to make a pure scanning workflow unworkable in almost all situations, including this one. Scanning is incapable of doing what modelling can (good topo, clean results, accurate representations of translucent, reflective, etc surfaces). Likewise, modelling is incapable of doing what scanning can (quick approximate real-world captures). They're different tools that accomplish different things. > In fact photogrammetry and laser scanning is getting to the point of eclipsing hand modeling for pretty much anything that is not a result of a CAD model. Due to differing definitions of what is considered "CAD" I'm not sure what you mean by this. I can think of applications where 3D scans are used, but only either as a reference for hand-modelling, or only after being so heavily hand corrected that it's a really more of a hybrid (model/scan) process. Animation almost always uses hand-modelled or heavily corrected (by hand) scans. Manufacturing uses hand-modelled objects or traced-over scans. Gaming uses hand-modelled objects or traced-over scans. Where are scans eclipsing hand 3D modelling? They're being used but as a different part of a hybrid workflow.
- AndrewKemendo 10y ago
  Well in our case we use models of furnishings. The space of furnishings is so large that you can't hand model everything manually in polynomial time (that's a joke). As a result we have clients that want 7 million objects in 3D. So we had to figure out how we were going to build a massive model repo quickly and at high enough resolution to be good enough for rendering. Scanning is the only way to do that so we built an entirely new way to do that combining the latest photogrammetry and depth scanning research. Going further, with already built things, that aren't previously built in CAD, like components and old buildings, in solidworks autocad etc... Making those models is much more efficient with structured light or photogrammetry systems. It should be noted that it's not just speed, it's hard to model to micron specificity, whereas we can with scanning. Animation is a little different but I think follows the same principles in the long run. The distinction between the methods in my mind is between digitizing something real versus making something new. I think in the former case scanning will win long term.
pen2l 10y ago
> In fact photogrammetry and laser scanning is getting to the point of eclipsing hand modeling for pretty much anything that is not a result of a CAD model. Woah, interesting. Please go on? (i.e. what would be your workflow for this - what applications/apps/hardware would you use for this? Also, for a thing with a lot of polygons, like a human, ok -- but would you do same for an object better suited for CAD programs? Like a monitor, let's say?)
- AndrewKemendo 10y ago
  There are a variety of them. As I mentioned elsewhere the core distinction is in digitizing real things that did not come out of a manufacturing process that utilizes CAD. For us, as we are an AR company, building 3D content is a critical piece so we need to digitize real objects as quickly and accurately as possible. The trick for us is finding the perfect balance between size of the resulting model and quality. So our scanning process is very refined.