Russel's teapot just got a lot more complicated.
I'll take your frame to another dimension, pay close attention: This AI auto-generates 3D objects from 2D snaps
Boring 2D images can be transformed into corresponding 3D models and back into 2D again automatically by machine-learning-based software, boffins have demonstrated. The code is known as a differentiable interpolation-based renderer (DIB-R), and was built by a group of eggheads led by Nvidia. It uses a trained neural network to …
COMMENTS
-
-
This post has been deleted by its author
-
-
Tuesday 10th December 2019 10:24 GMT Anonymous Coward
Re: re-write the heading
Ah, so that's been done for decades (literally, early 2000s and before IIRC).
There is an AI that generates from 1 picture:
https://gizmodo.com/nvidia-taught-an-ai-to-instantly-generate-fully-texture-1840323132
and:
https://venturebeat.com/2018/12/04/google-ai-generates-images-of-3d-models-with-realistic-lighting-and-reflections/
(Though I cannot find the one example I was thinking of. :P )
-
-
Tuesday 10th December 2019 10:44 GMT Anonymous Coward
+1 for the Prodigy reference in the title.
I have a friend who is involved in custom rc car chassis, and at the moment he says he has 2 months before the clones flood out of his latest design and kill the market for his new product, and then he starts getting complaints about quality and failures only to discover its badly made (as in materials of lower grade, less precise tolerances, cheaper electronics etc) clones. He is convinced he has two months lead time because his new designs are bought by someone with the intent to copy them, and that's the time to purchase & send the item, them to send it onward to the cloners to reverse engineer and production of the clone to start and be shipped.
If this takes root properly and its reasonably accurate enough to make a passable clone, they can cut out at least 5 weeks of that lead time just by using brochure information and publicity photos of the item.
-
Tuesday 10th December 2019 12:07 GMT Anonymous Coward
But creating such a thing requires replicating all the internal fixing points etc... which would not be discernible from a publicity photo. If the cloners are going to the lengths of adding those to the model themselves, they're basically doing the design work anyway so they might as well go the whole hog and start from scratch.
-
-
-
Tuesday 10th December 2019 17:15 GMT Anonymous Coward
great development
I have long waited for some automated process that would scan an old map (ok, at least an air photo), and turn it into a 3D (of course the buildings would be just a guesswork, from, say, a database of buildings created approx. in that time (and typical for specific country / area ;)
and it would be great to see automatically built, realistic 3d landscape in google earth at last! Then, combine this with gaming...