timzhang642
Bringing 3D ML to commercial CAD programs
Guys, with all the magnificent work done here is it possible to start bringing together the 3D ML and the traditional CAD programs (Creo, SolidWorks, Catia, etc...) This would be a real breakthrough in machine design world, reinforced with ML capabilities
- Geometry part/assembly classification and detection
- Identifying key characteristics of the model: size, volume, material, weight, etc.
- Parametric and geometric search
Currently there is no such capabilities in any commercial CAD program, and there's a great need for such application, especially in companies with a big 3D dataset, it will give a chance of smart search, based on geometry and CAD key properties, and ofcourse a major move towards design automation. As I see it the application shoud be integrated smoothly in currently commercial CAD program
It's possible that the accuracy is not high enough for commercial CAD applications. Right now object detection is reaching 90% accuracy. Maybe a few more conferences and some new improvements will come out to reach above 95%. Part segmentation accuracy is still below 88% on benchmark dataset.
I think it will happen in a year or so though. At least some proof of concepts tools, maybe integrated on the web (these ML methods are quite costly, computation-wise).
Hi @ Arthurg27 ,
How to train 3d Volume component ? How to extracting 3d volume features and how to matching this features ? How to segment 3d volume component based on 3d features ? How to calculate scaling of 3d volume components? How to classify the 3d volume components ? without GPU, how to train those above question ?
can you share steps of above question ? using dicom data sets?
Hi@selvakarna This article on medical 3D Image Analysis will give you an insight on what can be cone with brain MRI implementing 3D CNN
(https://resolver.ebscohost.com/openurl?sid=google&auinit=K&aulast=Kamnitsas&atitle=Efficient+multi-scale+3D+CNN+with+fully+connected+CRF+for+accurate+brain+lesion+segmentation&id=doi%3a10.1016%2fj.media.2016.10.004&title=Medical+Image+Analysis&volume=36&date=2017&spage=61&site=ftf-live)
Try this one as well: https://github.com/deepmedic/deepmedic
Hi @Arthurg27 , how to sign up above site ?
can you share ? dicom volume image training and 2d image based 3d volume segment ? CPU training link ?
It's possible that the accuracy is not high enough for commercial CAD applications. Right now object detection is reaching 90% accuracy. Maybe a few more conferences and some new improvements will come out to reach above 95%. Part segmentation accuracy is still below 88% on benchmark dataset.
I think it will happen in a year or so though. At least some proof of concepts tools, maybe integrated on the web (these ML methods are quite costly, computation-wise).
Any repos in particular that you know of tackling DL in the 3D/CAD space?
Guys, with all the magnificent work done here is it possible to start bringing together the 3D ML and the traditional CAD programs (Creo, SolidWorks, Catia, etc...) This would be a real breakthrough in machine design world, reinforced with ML capabilities
Hey, Arthur. Found your comment from three years ago. Today i think we are at a point where this technology can be developed. Infact I am investigating few of them.
