Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I don’t think the pervasiveness of ML articles on HN are an indicator of anything except hype trends on certain subject matters. ML research in these spaces has been very high output for many years now.

As someone in the field of computer graphics , where there’s been considerable ML research over the past few years that are more reliably applicable to people’s lives , most of the exciting stuff doesn’t make it to the front page of HN even if it’s posted here.

There’s been lots of research in the past few years. The initial shiny stuff makes it on here, but it’s the follow up iterations that are highly catalyzing of change that don’t because public interest in those topics has waned in the interim.



Would be great if you could share your 3-5 most exciting developments that weren't discussed mich here.


My mind might have quite a bit of recency bias since I just wrapped siggraph a few weeks ago (without catching Covid!) so I’m a little scattershot. I might come back to this later to post more concrete lists but here’s what came to mind first.

I should add that my point is more that there’s not been a slow down in ML research, and the ebb and flow of interest on HN isn’t indicative of accelerated progress in the field. Simply of what things have captured the public mind.

Anyway on to the links…

Disney had a slew of papers out this year, the facial motion retargeting ones in particular are very interesting for use in production of films and “metaverse” characters

https://studios.disneyresearch.com/machine-learning/

Luma have had a lot of progress in their Nerf capture and on device representation which will likely have huge effects for e-commerce use cases among other things

https://captures.lumalabs.ai/unbounded

Nvidia released an ML based version of OpenVDB that will potentially improve effects in films, but could be huge for games

https://youtu.be/uAs8X5es1DE

There were also a ton of neural rendering papers at siggraph that I still need to separate in my head mentally since I saw them presented back to back, so I apologize for just sharing a dump

https://twitter.com/neural_fields/status/1555947856271446018...

Apple released some neural rendering content too that has a lot of implications for spatial product training

https://arxiv.org/abs/2207.13751


You started with downvotes but showed up with the links. Neural VDB looks wild - I can see games going nuts with detailed, interactive volumetrics now, with such a tiny memory footprint.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: