Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

Предыдущая публикация
tctc

I mean, yeah, I say "simulacrum" because it will not be an actual item belonging to the dataset. That's why the best cases of the technology do not require the original to be reproduced, where tolerances are large enough an approximation serves just as well

110
tctc

for example I think it looks kinda shit but the neural textures nvidia demoed for displaying extremely intricate items like jewels with internal reflections is an application that makes sense. It's not necessary to reproduce an actual, specific sapphire from the dataset. "A sapphire" will do

Обсуждение

1 прямых ответов · 3 сообщений

まひろ・アポカリプス

that is not what I meant I mean that you cannot reproduce the same, or even close to the same, non-original sapphire as you got in attempt 1

Ответ для まひろ・アポカリプス
まひろ・アポカリプス

it is possible only as long as no parameters are changed, and no randomisation is applied to weights

Ответ для まひろ・アポカリプス
tctc

like I said, a specific sapphire doesn't matter, it doesn't even matter if the same sapphire is produced every time. It just needs to be "a sapphire" But also yes for something like neural textures they wouldn't be randomizing weights. Things are being produced based on the "ground truth" anyways