1 2
Posted: 12/27/2016 4:29 PM
Quote
Ramzi
Joined: 3/25/2004
Posts: 459
If a human can do it, and AI eventually can. Maybe one day we'll see such gorgeous remakes made easily.
Posted: 12/27/2016 4:48 PM
Quote
ThunderAxe31
Any
Editor, Reviewer, Skilled player (1359)
Joined: 9/12/2016
Posts: 1646
Location: Italy
Arcade TAS of 2023 Homebrew TAS of 2023 PSX TASer of 2022 Gameboy TASer of 2017
Ramzi wrote:
If a human can do it, and AI eventually can. Maybe one day we'll see such gorgeous remakes made easily.
Maybe it is possible to create a batter depixelizing algorithm, but that would still not be enought to remasterize a game. Remastering games need arbitrary choices that can't be automatized in any way. It needs a lot of personal choices and artistic sensibilities... that's what the "remastering" word itself means. That's why remastered games needs so much work in order to be made.
my personal page - my YouTube channel - my GitHub - my Discord: thunderaxe31 <Masterjun> if you look at the "NES" in a weird angle, it actually clearly says "GBA"
Posted: 12/27/2016 4:54 PM
Quote
Ramzi
Joined: 3/25/2004
Posts: 459
And some day and AI will be able to make those personal artistic choices.
Posted: 12/27/2016 11:23 PM
Quote
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
Ramzi wrote:
And some day and AI will be able to make those personal artistic choices.
We're decades to centuries away from being able to make an AI with a personal aesthetic taste (rather than just being able to data mine and make a neural network that creates an average image of something, for example), so don't hold your breath on it.
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
Posted: 12/27/2016 11:35 PM
Quote
ThunderAxe31
Any
Editor, Reviewer, Skilled player (1359)
Joined: 9/12/2016
Posts: 1646
Location: Italy
Arcade TAS of 2023 Homebrew TAS of 2023 PSX TASer of 2022 Gameboy TASer of 2017
Ramzi wrote:
And some day and AI will be able to make those personal artistic choices.
If you let a machine take that sort of choices in your place, then life stops having meaning. If for example, someday there would be created an AI than can do perfect TASes, then human TASing would lose much meaning. One of the good things abot TASing is that you can't never be absolutly certain that a run can't beaten. That's what makes you give it all to desperately search for a way to beat current records.
my personal page - my YouTube channel - my GitHub - my Discord: thunderaxe31 <Masterjun> if you look at the "NES" in a weird angle, it actually clearly says "GBA"
Posted: 1/27/2017 6:03 PM
Quote
WST
She/Her
Active player (490)
Joined: 10/6/2011
Posts: 1697
Location: RU · ID · AM
Love this example
S3&A [Amy amy%] improvement (with Evil_3D & kaan55) — currently in SPZ2 my TAS channel · If I ever come into your dream, I’ll be riding an eggship :)
Posted: 1/27/2017 10:07 PM
Quote
Archanfel
Skilled player (1458)
Joined: 11/26/2011
Posts: 656
Location: RU
Sega TAS of 2018 Sega TASer of 2015 Sega TAS of 2015
Patashu wrote:
We're decades to centuries away from being able to make an AI with a personal aesthetic taste.
Not be so pessimistic at forecast. I remember a just a few years ago everyone said that we're decades to centuries away from being able to make an AI that able to defeat best human Go players. However Alphago devs has disagreed to wait so long. Link to video Nowadays it has 60-0 win streak to best human Go players. So may be we are already close enough to the Day when self-learning AI which are smarter than humans in many other fields will be created.
I show you how deep the rabbit hole goes. Current projects: NES: Tetris "fastest 999999" (improvement, with r57shell) Genesis: Adventures of Batman & Robin (with Truncated); Pocahontas; Comix Zone (improvement); Mickey Mania (improvement); RoboCop versus The Terminator (improvement); Gargoyles (with feos)
Posted: 1/27/2017 10:18 PM
Quote
feos
Site Admin, Skilled player (1254)
Joined: 4/17/2010
Posts: 11486
Location: Lake Char­gogg­a­gogg­man­chaugg­a­gogg­chau­bun­a­gung­a­maugg
NES TAS of 2011
Archanfel wrote:
So may be we are already close enough to the Day when self-learning AI which are smarter than humans in many other fields will be created.
http://tasvideos.org/forum/viewtopic.php?p=445643#445643
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
Posted: 1/27/2017 11:17 PM
Quote
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
AlphaGO is super cool and a massive achievement, don't get me wrong. But it is still categorically different from an AI that can learn in a generic sense (rather than being tailor made for a certain game or simulation), especially one with emotions or biases.
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
Posted: 1/27/2017 11:45 PM
Quote
Archanfel
Skilled player (1458)
Joined: 11/26/2011
Posts: 656
Location: RU
Sega TAS of 2018 Sega TASer of 2015 Sega TAS of 2015
Patashu wrote:
But it is still categorically different from an AI that can learn in a generic sense.
Yeah it true, all modern AI are very highly specialized. However it does not mean that this situation will last forever. Sooner or later breakthrough will happen. It just a matter of time. I not too naive to say that it happen soon. But if you ask me will it happen before 2042? I say "yes" is much more likely than "still no".
I show you how deep the rabbit hole goes. Current projects: NES: Tetris "fastest 999999" (improvement, with r57shell) Genesis: Adventures of Batman & Robin (with Truncated); Pocahontas; Comix Zone (improvement); Mickey Mania (improvement); RoboCop versus The Terminator (improvement); Gargoyles (with feos)
Posted: 1/28/2017 2:48 PM
Quote
BadPotato
Joined: 1/26/2009
Posts: 558
Location: Canada - Québec
This project use a DNN to upscale 16x16 images by a 4x factor. So the output is 64x64: https://github.com/david-gpu/srez Similar experiments could be made with low resolution games. That being said, I'm not sure what could be used to train a model for these games. Maybe feeding random NES games picture as input and expecting for random SNES quality picture as output could lead to something interesting.
Posted: 1/28/2017 3:20 PM
Quote
Aktan
Publisher
Joined: 4/23/2009
Posts: 1283
This is borderline AI, I think, as it uses neural networks to figure out what looks best for humans: Link to video It's using an upscaler caller nnedi3.
1 2