ON 20230101@4:33:39 PM at page: On a web page you were interested in at: http://techref.massmind.org/Techref/method/ai/ConvolutionalNeuralNetworks.htm#44927.6900347222 James Newton[JMN-EFP-786] Says <P> This talk by Vincent Sitzmann of CSAIL <BR><A HREF="https://http://www.youtube.com/watch?v=Or9J-DCDGko">youtube.com/watch?v=Or9J-DCDGko</A> had a very interesting tidbit near the end, where he showed a VERY well captured 3D scene of a room which was encoded in a 1MB (!) set of neural network weights. I was sort of blown away at that image compression. Basically, the NN had learned to reproduce the 3D scene independent of voxel resolution. Obviously, the training time for that would be prohibitive for real time operation, but as a way of transferring the initial 3D scan of the work area, it might be quite useful. <BR><A HREF="https://vsitzmann.github.io/">vsitzmann.github.io/</A> "Implicit Neural Representations with Periodic Activation Functions" <BR><A HREF="https://vsitzmann.github.io/siren/">vsitzmann.github.io/siren/</A> The specific section on "learning" 3D structure from a point cloud is: <BR><A HREF="https://github.com/TalFurman/Implict_neural_representation_of_images#sdf-experiments">github.com/TalFurman/Implict_neural_representation_of_images#sdf-experiments</A> <P> Sample code in a Google Colab (free online) is here (under Community...) <BR><A HREF="https://arxiv.org/abs/2006.09661">arxiv.org/abs/2006.09661</A> Sadly, the 3D example is not included. <P> More in-depth discussion of the method: <BR><A HREF="https://http://www.youtube.com/watch?v=Q5g3p9Zwjrk">http://www.youtube.com/watch?v=Q5g3p9Zwjrk</A> ON 20230101@5:55:31 PM at page: On a web page you were interested in at: http://techref.massmind.org/Techref/method/ai/NeuralNets.htm# James Newton[JMN-EFP-786] edited the page. Difference: http://techref.massmind.org/techref/diff.asp?url=\Techref\method\ai\NeuralNets.htm&version=0 ON 20230122@9:57:55 PM at page: On a web page you were interested in at: http://techref.massmind.org/Techref/idea/games.htm#44948.9152199074 James Newton[JMN-EFP-786] Says <H2>Procedural Generation</H2> <P> Using pseudo random sequences, a single seed value can be used to calculate the same sequence of values every time. If these values are tied to the generation of game elements, each seed can generate an entire world, of extreme detail. <P> Each value can be taken as a seed for a subsection of the game environment. In this way, sections not currently in use can be ignored, and only the seed for the current section unpacked. <P> The only difficulty here is that one must unpack all the values to get to the one you need, as they must be processed in order each time. Algorithms like the <A HREF="https://en.wikipedia.org/wiki/Bailey%E2%80%93Borwein%E2%80%93Plouffe_formula">Baileyâ€“Borweinâ€“Plouffe formula</A> might provide relief from this requirement.
|file: /Techref/new202301.txt, 2KB, , updated: 2023/1/22 21:57, local time: 2023/1/27 06:16,
|©2023 These pages are served without commercial sponsorship. (No popup ads, etc...).Bandwidth abuse increases hosting cost forcing sponsorship or shutdown. This server aggressively defends against automated copying for any reason including offline viewing, duplication, etc... Please respect this requirement and DO NOT RIP THIS SITE. Questions?|
<A HREF="http://www.piclist.com/techref/new202301.txt"> new202301</A>
|Did you find what you needed?|
PICList 2023 contributors:
o List host: MIT, Site host massmind.org, Top posters @none found
- Page Editors: James Newton, David Cary, and YOU!
* Roman Black of Black Robotics donates from sales of Linistep stepper controller kits.
* Ashley Roll of Digital Nemesis donates from sales of RCL-1 RS232 to TTL converters.
* Monthly Subscribers: Gregg Rew. on-going support is MOST appreciated!
* Contributors: Richard Seriani, Sr.
Welcome to www.piclist.com!