I really enjoyed this blog style. The code stuff is cool but hearing your thought process and getting deeper into the concepts like this is far more valuable imo
> "maybe in a few years I can just raytrace" HOO WEEE BOY, God I know that one all too well. I did a similar thing. I was working in reinforcement learning and there were some key problems, so I thought "oh you know I'll go do some work in supervised learning, hopefully by then the problems are alleviated". So a year goes by and the field has improved somewhat, but now the mindset is shifted toward mega-compute. So I think to myself, "okay well ill go spend the time in pure mathematics, and hopefully by the time I get back things will be more stable ...".
This is absolutely great stuff! You're work resonates to me incredibly! I'm both inspired and humbled by this type of effort. To be working on so much independently is phenomenal!
TheCherno : Explains shadow that .. shadow these.. Meanwhile Me Who doesn't even knows how to print something on the console : hmm...I see so that's how it is..
This is a great vid, I'd love to see more general videos like this where you pick a topic and talk briefly about different implementations and their tradeoffs. It saves me a lot of research!
Nice ! - adding shadows really makes the scene look so much better. Your engine is coming along nicely - congrats. My skill set is very low in game engines so most of your stuff is way above my understanding but can see all the hard work you have put into it at least...making my first game currently using an old game engine on it's last legs I guess but not terrible either (CopperCube 6) as I originally didn't want to learn to code at all....funny how that never eventuated. I now do 99% of my game in Javascript now...erhummm...
Shadows are such a fun and frustrating topic. A few alternative/complimentary techniques to shadow maps are screen space shadows which can be combined with shadow maps to enhance small details and soft shadows using signed distance fields. SDF shadows are used by the unreal engine to shadow objects outside the shadow map range. An interesting enhancement for cascaded shadow maps is a tight fit to the depth buffer range. This can be achieved by creating hierarchical depth containing the depth min/max. The light matrices can then be calculated in a compute shader to keep everything on the GPU. I think the Order 1886 used this technique. Shadow map sampling also has a bit of varierty. Besides the traditional "blocky" PCF sampling a lot of people use per pixel rotated poisson disks. An alternative is to use a spiral pattern as those have less sample overlap when rotated. And of course going fully stochastic with a low sample count and cleaning up with TAA. Another interesting approach is using screen space directional occlusion to approximate short soft shadows. Although I believe this is mainly used to provide shadowing for lights that can't afford a shadow map. This was originally used in Cryengine 3 and more recently in the Demon's Souls Remake.
Just came here to thank cherno for everything he is doing . His C++ series and OpenGL series is what I am currently learning. Once I start earning I will surely contribute to Cherno . I love your efforts Cherno. I am currently 21 and love programming. Once again thanks and love from Uttrakhand , India.
Really sleek! I'm not close to being a game engine developer, but implemented shadow mapping a few months ago nonetheless. It took me quite a while and the shadows look very rough and overall just not polished but it was really fun and it makes all the difference when you're looking at your scene during development and it has shadows vs everything's kinda flat ;)
About what you said in the beginning. I find working on code and pretty much any other creative work at the same time to be very hard. Let's say the brain is a workshop, once I've put out all the tools needed for coding, there's no space left for the tools needed by others type of jobs and I have to put all your code tools back in the cupboards before starting another job. Otherwise it's too messy to make sense of anything. But I know someone that doesn't seem to have this issue.
For debugging things like cascades (and many other things in fact), you should have two types of cameras. 1) The game camera, that is what your rendering is based, all the frustrums, cascades, culling, LODs, etc. 2) Debug / editor camera, which allow you to actually move around, and see things from different viewpoints, without affecting any. If there are some shaders that take into account some view vectors, you might have additional option which camera to use.
I would say best solution for implementing some algorithm would be to mostly use the paper, and use some "random" code from github for reference if/when you you get stuck from the pure paper (which will happen). Haven't done this a lot for 3D graphics algorithms (yet), but it was my goto when doing more advanced AI algorithms in uni.
9:39 (ish)... I hope when you are copying open source code into your own (not open source) code you are complying with the license. A quick search on github I found both things that are ok (mit och apache 2 for example) but also many things that are not (no license)
Hi Cherno, How did you make this engine? I mean how did you start? You just made a file named hazel.cpp and started coding ? or you're using something like Qt Creator for the GUI?
Great! I would like to ask if there is idea of adding reflection to C++, and what other main functions of reflection in game development besides serialization and development of engine editors. Thanks!
@@__jan that means we have to wait for 3 years at least...I.ve write an static reflection for cpp....But I don't know the specific usage requirements of reflection in game development to improve it.
@@ventusstrife6715 One use case that I use it for is scripting language binding. If you can bind member functions to a scripting language and create an API with that it'll drastically increase the flexibility of the engine. I did mine for Lua binding, and the method that I used was I defined functors to pull individual values off the stack, and called the member function with std::apply and an array of those functors. It's quite a bit more involved than member/data reflection, but it's well worth the investment.
The first time I tried to implement shadowmaps into my custom DX12 game engine, I tried to store the back buffer render target into a variable and re use that variable every frame, not even considering the fact that double buffering will mean that my variable is the wrong texture view pointer on every other frame. This lead to DirectX throwing a fit with unreadable _com_error exceptions, and me spending an entire day hunting down the problem. Not a fun experience. These are the things that get game developers. Trying to do complex stuff is not the problem. It's making stupid mistakes while in the process of trying to do the complex stuff. Anyway, I'm still in the process of implementing my shadowmaps, and just dealing with the shadowmap "view projection" transform is tough enough. In my engine, each visible room can have its own shadowmap transform, and I have to combine them all together into an overhead view that projects onto a single depth texture, where closer rooms are rendered at a larger scale. So I have to implement some form of "uv packing", where each room acts as a separate island. Lots of respect for implementing this into a custom game engine. It takes some work.
Idk if this already exists. But, they should make a method where it combines shadow mapping and ray tracing. So for complex parts of shadow use ray tracing and use shadow mapping for other parts.
I had idea 1: You approach half open door it is light, with young people: meaning the leaders talk behind doors high loud bouncing sound. 2: The dark: a half open door leading to the cellar, ok you go down there have to light a torch still walk down the at far se a room with light coming out sound like merry go around sound really scary should you go there or not?! If look at the two images (world leaders, contra the center of the earth) interesting imaginative image
I like the style and length. maybe a bit too much talk in the beginning, I would have loved to see some of those other github projects you tried and what was wrong with them. Keep up the great work!
Hey, very cool video. I was just curious about something as I tried to implement PCSS a little while back, and I would have issues where if one object was directly in front of another (like a pole in front of a wider pole) with the light facing them dead-on, the blocker search would give a value closer to the front pole's depth when sampling for points that are behind them both, and the resulting shadow would have a high-penumbra streak in the shadow of the pole in the back. So essentially there would be like an inverted shadow of the front thin pole behind the back wide pole. Tweaks to the numbers for blocker search size and everything could fix for very specific cases, but never for general cases. Did you ever encounter this issue, and if so how did you work around it? Do the cascades fix this issue, as I did not get around to implementing those before dropping for variable penumbra for the time being, in favor of another solution (currently using prefiltered EVSM and it looks really nice, but that contact-hardening is still so nice).
Hello cherno, i started play NFS No Limits game from 2018.i see your name "Yan Chernikov - Game Engine Team" in the credits.btw, that game has a really good graphics for android and ios game compare to grid 2 android port.
Are you also going to implement shadow maps/baked lighting? I imagine that as long as one doesn't do something as silly as dynamic time of day, this could be a giant win in performance and visuals. It would probably need to be ray traced with something like Radeon Rays or something. I imagine Cinema 3D would be cost and license prohibitive.
i know nothing about game development but i just like this kind of stuff i do understand some of the stuff you're talking about but sometimes i go like what the heck is he talking about i don't know but it's interesting
To me it looks like the shadows are only softened in one axis (towards the light) is that for performance reasons? The shadows look really good though. I think the next step would be to integrate some nice Ambient Occlusion/GI.