วีดีโอ

At 40:30, the low frequency noise in the bottom left image looks similar to the "boiling" artifact which denoisers often produce in ray traced games. Does this suggest the boiling is caused by the usage of white noise? Another question, may spatial or spatio-temporal blue noise also be used to improve diffusion models? I looks like they are using white noise currently.

you mean high framerates

That filtering point of view was very cool to see. Neatly explains why blue noise looks clean when viewed farther away. Looking forward to learning more about FAST noise as well!

Looks insane, thanks for the talk

Great overview! The story of loot boxes reminds me of an article I read in Game AI Pro. In that article, they also found that players perceived pure random samples as "unfair." What that article did was reject certain samples if the same result was produced repeatedly too many times.

Fantastic talk, thank you!

Great stuff! Enjoying using GIGI for rapid prototyping of some ideas I've wanted to try for a while. More videos like this please :)

While it is stable, is it efficient enough to be used in games, perhaps if it were to be able to take advantage of the less used 2 cores in the most common 8 core CPU’s just curious, thanks for your explanation

These vids popped into my head recently … from back when AI weren’t in the normie zeitgeist. I saw one where you could train AI to help make maps but don’t remember if it was SEED, I remember the vid as having cute bushes, vaguely cartoonlike.

This is awesome, I can already imagine using this for prototyping various kinds of graphics techniques.

вы продаете или только показываете?

This is really COOL! Anybody you know of make a blender implementation for this yet?

MPM is the most fun class of simulation that exists imo. Would love to see this in games someday!

I'm puzzled why profit-hungry short-sighted companies like EA permit their researchers to publish results publicly instead of patenting and/or using it for competitive advantage.

This looks like it'd be a cool approach for a plate tectonics simulation

Very interesting to see the physics going into game development.

really awesome video! thanks for sharing

I am a touchdesigner user and i aim to use this for some interactive installations… keep up with this page ❤. You push me to become a better artist with all the knowledge you share freely here. A big thanks

Really awesome stuff. Ill defo love working on this.

Wow, your videos are the right stuff to get even more interested in game engine development. Thank you and greetings from the Walrus :D

This seems pretty fire 🔥

My friend at EA just introduced me to this, seems like an excellent tool for prototyping a number of things I've been working on in the past!

This looks cool, but sadly it doesn't build straight from Github. Missing ImGui library I think.

hopefully vulkan support will be implemented soon or released if it already implemented.

Let's goo

cool to see my old slide in the beginning! i have another anecdote about the benefits when using TAA the application of blue noise on reflections just came about from seeing a high quality image on the "dithering" wikipedia article applying it i noticed that it looked a little better than white noise without TAA, but with TAA the difference was much bigger the negative correlation was specifically helpful while using TAA specifically because of TAA's neighborhood clamping/clipping essentially it ensured that the neighborhood AABB would always be big enough to always contain a new sample, avoiding rejection whereas with white noise, often times TAA takes in the reprojected sample, sees that it's outside the random box, and tosses it gonna check out that FAST, now that the i3d session is online, thanks for the cool presentation!

Blue noise textures looks suspiciously similar to Turing patterns

Oculus has had libraries to do this for 5+ years

thank you youtube recommendation algortihm. my brain is smarter now

I would argue that you can't have a random function that is completely fair, because it would need to be deterministic and random at the same time. that's what I get from this video.

Excellent presentation!

I really learn a lot from this video, thank for the quality content! I didn't know that you can optimize the noise pattern based on different filter. I was wondering from a computer vision perspective, we can use back-propagation to learn good convolution filter. But in your case, given a set of image samples, what if you optimize it stage? In other words, learn best filter, and then learn best noise, will it improved?

This is/ was a great video; worth my time.

At 4:00 the answer would realitically be somewhere in between 50% and 100%. Much closer to 50% than 100%, but the probablility of the coin being unfair is to be considered. If we estimate the prior probablility of the coin being unfair (landing 75% of the time on one side) to be 1% for each side, after 3 consecutive heads the probability of the coin being biased towards "heads" raises to 3.3% and the probability of it being biased towards "tails" drops to 0.1%, making the overall probability of getting "heads" next time to be slightly above 50.8%.

Great presentation!

I wonder if a natural-looking prior or a collected statistics on the level of neighboring pixel's spatial correlation can be used to construct a better noise (instead of using assumption of small delta x to small delta y).

Wolfe, Thank you for the great video. I've learned a lot from your blog and I'm glad to be able to express my gratitude on this channel.

"RNG" in gamer talk doesn't mean "random number generator", just inherent randomness or your luck with it. It's not that gamers think the generator is bad when they have bad luck, they just use the acronym as an expression for something else. "Bad luck with the RNG" got shortened to "Bad RNG" but meaning the same thing.

Holy smokes, EA can't afford any audio production? Are you calling in from inside a sea can?

This is one of those videos I wish I could give more than just one thumbs up. Very well presented.

I would argue that from both the implementation and gameplay perspective, keeping loot tables random is the best choice. While you do "smooth out" the gameplay experience, you also lose out on more personalized experiences, for example a player having a lucky day and getting many rare drops in a row.

Blue noise rendering kinda looks like analogue photography. So cool!

Nice talk! I liked the visuals, I think that's the most colorful Rendering Equation I've ever seen!

3:40 The thing with Gambler's Fallacy is that, statistically, it WILL WORK for a PART of the people. If you look at e.g. 1000 people each doing 4 coin flips you expect "There's no way I get 4 Heads" to hold for 937 of them (=1000*(1-0.5^4)) and about 60 of those will get the HHHT falacy scenario P.S. You can run the trials with a 1-liner in your browser's console: [...Array(1000)].map(person => [...Array(4)].map(isHead => Math.random() >= 0.5).every(isHead => isHead == true)).filter(allHeads => allHeads == false).length

Why is this one hour long when it can be explained simply in 5 minutes 😑 You don’t need to storytell the whole history of computer graphics and math to explain a simple RNG 😑

Very interesting video. Well done.

Great presentation! Do you think that blue noise is relatively easy to implement in softwares retrospectively? I am sure that adding blue noise systems moving forward will be more popular.

Thanks!

3:32 Isn´t it basicly saying that no previous outcomes have an influence on the next result. The odds are ALWAYS 50/50? I had 20 times black in a row in a roulette game once. Lost me a good chunk of money that evening 😀

From electronic arts, the randomness is based on the money spent on DLCs