search

w3c / machine-learning-workshop

Site of W3C Workshop on Web & Machine Learning
https://www.w3.org/2020/06/machine-learning-workshop/
42 stars 12 forks source link

WebGL garbage collection #63

Open dontcallmedom opened 4 years ago

dontcallmedom commented 4 years ago

@jasonmayes' talk and @annxingyuan's talk mention the challenge of garbage-collecting with the WebGL API in the context of providing a library on top of WebGL-managed resources.

@grorg is there a well-known approach to deal with this, or ongoing work in WebGL land to address this? Is this also an issue in the context of WebGPU?

jasonmayes commented 4 years ago

I mentioned my thoughts here on the other thread #66 related to WebGPU. In essence most JS devs do not think about memory cleanup as the JS garbage collector works so well, but as ML becomes more popular these sorts of leaks will certainly grow. This can help not only ML model execution from springing memory leaks but also folk working with graphics generally speaking too.

Adding @pyu10055 @dsmilkov @nsthorat @annxingyuan @tafsiri @lina128 for any thoughts related to this topic too (TF.js team) and for visibility.