Sam Bigos

562 posts

Sam Bigos banner
Sam Bigos

Sam Bigos

@Calneon

It's dangerous to stay a while and listen.

Brighton เข้าร่วม Ağustos 2009
289 กำลังติดตาม740 ผู้ติดตาม
Sam Bigos รีทวีตแล้ว
Assassin's Creed
Assassin's Creed@assassinscreed·
Watch the Gameplay Overview of #AssassinsCreedNexusVR now! Check out more than 8 minutes of new gameplay footage and learn everything you need to know about combat, parkour, stealth and accessibility features Assassin's Creed Nexus VR is coming November 16 to Meta Quest 2 and 3
English
23
74
557
114.8K
Sam Bigos
Sam Bigos@Calneon·
@gadirom_ @ziyangwen It's the one you linked in the post you sent me (the browser game, I don't know if the renderer is released). Normals and depth exist in 2D. But I don't know enough about how IS works. I assume it would work similarly.
English
1
0
1
15
Roman Gaditskii
Roman Gaditskii@gadirom_·
@Calneon @ziyangwen Not sure how IS works in 2D, also denoisers usually asks for normals and depth which don’t exist.. can you give me a link on his renderer?
English
1
0
0
18
Sam Bigos
Sam Bigos@Calneon·
@gadirom_ @ziyangwen Ah nice. Yeah mine is very slow too. I think you could use some importance sampling techniques that 3D GI solutions use to make it faster. Impressive how fast kishimisu's browser version runs!
English
1
0
1
40
Roman Gaditskii
Roman Gaditskii@gadirom_·
@Calneon @ziyangwen Here is how my implementation looked like. Extremely noisy and quite slow on iPhone. I used a lot of Gaussian to fight noise but then the detailed struggled.
Roman Gaditskii@gadirom_

I saw the game by @kishimisu and the idea stroke me that it’s possible to smoothly blend between the integral of light transport function and the solution of Laplace equation by blending between normal distribution and singularity in WoS algorithm.

English
1
0
0
84
Roman Gaditskii
Roman Gaditskii@gadirom_·
@ziyangwen Wow Unity has a real-time 2D GI? I wonder what algo they used, since my Monte Carlo estimator looks ugly compared to this 🙈
English
1
0
1
67
Sam Bigos
Sam Bigos@Calneon·
Sci-fi shooty shooty I'm prototyping. Uses a global illumination system I wrote based on DDGI, and the character anims are all procedural IK (Fabrik).
English
1
0
9
804
Sam Bigos
Sam Bigos@Calneon·
@reduzio Hi Juan, are you able to give more info on how the solid/emissive/albedo maps are created? I'm not sure on the step to go from 2D rasterized images of the scene to a 3D texture of voxels. Is Voxelization the technique used here?
English
0
0
0
78
Oskar Stålberg
Oskar Stålberg@OskSta·
So.. you absolutely have to flatten all data structures to fit them into Unity's Jobs and Burst, right? My WFC library is really quite nested, so this looking like a real piece of work...
English
9
0
37
0
Sam Bigos
Sam Bigos@Calneon·
@yaqwsx_cz The very odd thing is that the warping comes in pairs, when one warps, it warps its adjacent neighbor in the opposite direction. I guess that's because when it warps, it then creates turbulence in the resin that warps neighbors? Tried adding rest time but didn't help much :(
English
0
0
0
0
Sam Bigos
Sam Bigos@Calneon·
@yaqwsx_cz Should be perfectly straight. My best guess right now is it's some external vibrations, possibly from the downstairs neighbor. It's inconsistent, some have worked, and when it warps the height varies. I agree doesn't seem mechanical.
Sam Bigos tweet media
English
1
0
0
0
Sam Bigos
Sam Bigos@Calneon·
@yaqwsx_cz Hi! Your post on resin printing are amazing and I'm learning so much from them. In a video (youtube.com/watch?v=TIYiXM…) you're printing tall and thin objects. I'm trying something similar but mine get warped at various heights. Did you face similar issues? Any ideas?
YouTube video
YouTube
Sam Bigos tweet media
English
1
0
0
0
Sam Bigos
Sam Bigos@Calneon·
@SebAaltonen Yes I'm talking about main thread work too. We make sure they don't hitch (are consistent) and are no more than like 1% of the total available frame time (e.g. if we have 5ms for game logic per frame, each task is no more than 0.1ms). Then, it averages and total variance is low.
English
0
0
0
0
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
@Calneon If these are background tasks which run in worker threads and don't stall the frame, then it's OK. I am talking about stuff that must be executed as part of the frame, and can cause a hitch -> miss vsync.
English
1
0
0
0
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
The most malicious performance hack: You have slow running code. Let's execute it every other frame (or every second in case of online services). The code is still slow, but is less visible in the average fps counter. But you will see horrible stuttering instead.
English
6
2
74
0
Sam Bigos
Sam Bigos@Calneon·
@SebAaltonen It can be the only way to improve performance if that code is already well optimised, though. If enough systems use it, and you have a reasonably intelligent scheduler, it averages out and saves a lot of time for systems that can handle it (we use 3hz and 1hz groups).
English
1
0
0
0
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
This is especially bad if multiple people add code like that and they amplify each other. The stuttering becomes bigger. It's difficult to profile performance issues like this as they are not visible every frame. You need to carefully time your frame capture.
English
4
0
22
0
Sam Bigos
Sam Bigos@Calneon·
@beast_pixels Yeah that's basically what the top image is. WFC isn't doing much except placing the correct tiles for the path. But, any other solution without WFC would encounter the same problem, making a path that doesn't make best use of the tiles, or allow weighted placement.
English
0
0
0
0
Sam Bigos
Sam Bigos@Calneon·
Applying path constraints to my marble run generator is a tough problem. Generating a path first then constraining WFC by it (top) ignores weights and biases inherent in tile setup. Leaving WFC to generate a path (bottom) leads to invalid paths and loops if using forked tiles.
Sam Bigos tweet media
English
2
0
6
0
Sam Bigos
Sam Bigos@Calneon·
@runevision I agree it can sometimes be a trap. But I think it is helpful here. I need to get from a start to end cell, doing that by randomly placing tiles wouldn't work. If you predefine a path, you're basically back to the initial problem where the path doesn't make good use of the tiles.
English
1
0
0
0
Rune Skovbo Johansen →🦋🐘
@Calneon If it's proving this difficult, I suspect it might be easier to use a non-WFC approach instead. Like continuously placing new valid adjacent tiles next to open ends, and backtrack if it's built itself into a dead end.
English
2
0
0
0
Sam Bigos
Sam Bigos@Calneon·
@OskSta @ValakhP Yeah I really like how that works in Townscaper. It looks like you're somehow collecting faces from the constructed mesh, constructing a new mesh and enlarging it slightly. Doesn't seem to be based purely on the face normal though since details like chimneys are preserved.
English
0
0
0
0