Posts for Selicre

Posted: 12/5/2019 7:20 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
r57shell wrote:
You said they are hidden: I was planning to do the same, but... Also, fix this stuff. https://imgur.com/a/emK20jl
I'm not actually sure what the cause of that is. It might be that the compressor read the state file before it was fully written. There's not a lot of these, but I might have to re-run it again, making sure that the filesize is correct before reading.
r57shell wrote:
Btw, with rust you could write plugin instead of my dll, because rust is c++ friendly.
I don't want to interact directly with Gens because I am running it in wine, and.. building 32-bit windows executables? That link to OpenGL? On Linux? No thanks. Also, that lets me fast-forward much faster because qbsdiff decompression is a ton faster than emulation.
Posted: 12/5/2019 6:34 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
r57shell wrote:
I tried camera smoothly following both, and with background enabled it makes weird move of it. I have issue with tails when he is "dead" because I don't know how to know is him alive atm, and I don't want to spend time on it. But when his position just reset - camera works fine. Just when he is dead, it's trying to reach some offscreen position.
I just set tails' position to sonic's position when tails is either at 0/0 or has x > 0x7000. Good enough for me. The issue with displaying the background together with the smooth camera is that the BG is inherently tied to the current camera position. If I were to try to display the background I'd have to emulate the behaviour for all animated tiles, which is not something that I have the time for. The glitchy (e.g. boss) sprites are still there, but they're not shown: I'm reading the object list at a specific point in time, when sprites are drawn: https://gogs.selic.re/x10A94/atlas-s3k/src/master/dumpframes.lua#L26 - this also means that lag doesn't screw up how sprites are drawn, since they always use the relevant info rather than "next frame" info. I'm currently experimenting with showing a short range of objects with known mappings. In the final encode, they will likely be shown at all times. The only problem I have is that the respawn table data doesn't match up with the object data, so there's a slight discontinuity. This'll likely require me to re-bake all savestates (which, btw, I've delta-compressed with qbsdiff to take up around 400MB rather than 209GB).
Posted: 12/4/2019 7:27 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
boct1584 wrote:
Selicre, I was specifically thinking of showing the camera boundary, but a gradient shading extending out from where the camera boundary is might work too. My concern with something like that is that it might make Tails's offscreen shenanigans hard to see.
Oh! I thought you were referring to the world boundary, i.e. the death plane and the camera scroll stop points. The reason why I didn't add this is because of how frantic the screen actually is; you can see that with e.g. the score tally in Ice Cap 2. It makes things a bit distracting. I'll definitely draw it when you are sufficiently offscreen, though.
Posted: 12/3/2019 6:44 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
Thanks for the kind words!
boct1584 wrote:
Could you reasonably add a border graphic to show where the screen boundary is? I don't think it's critical, but it would be nice.
So fun thing is, I kinda didn't wanna add it because as of right now the entire viewport is drawn with the same GLSL shader - e.g. everything including the menu is drawn using paletted tiles. Adding something like a border would require me to write a new shader, which - it's a good idea! I just wanted to push it out as fast as possible. How do you think I should draw it? Fully shade the unreachable area, or, say, use a gradient line? I've been thinking of also drawing the camera boundary if sonic was sufficiently outside of it.
Posted: 11/29/2019 5:15 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
Link to video So this is a full rewrite to make this thing realtime. I've changed a bunch: * Plane B priority tiles are shown when necessary * Added a HUD showing some technical info and ring counts * Made the camera movement interpolated - there are some silent teleport issues but I know of a workaround * Objects no longer flicker when on screen * The camera is no longer constrained to level boundaries and is able to show objects outside of levels - doesn't render loopback tiles yet, but I might change that * Removed the checkerboard background because it unsurprisingly murdered the bitrate Things left to do: * Render out-of-bounds chunks when necessary * Splitscreen for tails when the distance between S&T is too big * Account for compound sprite X overflow a bit better * Maybe use vertical offset-per-tile data? I don't really know how I'd implement that * Make the final boss and ending not horrifyingly broken
Posted: 10/31/2019 12:11 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
feos wrote:
Selicre, would it be feasible to have at least some version of the full encode done until December? We'll need to handle this submission until New Year.
Sorry about the delay! Real life stuff got ahold of me. I'll try to work on it this month; I feel like even if it's not perfect, it'll be received well enough that we don't need to bother with all of the edge cases, and I'd be able to make a better version if I have the time and energy to put into it.
Posted: 6/27/2019 6:23 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
feos wrote:
Dumping at 2x resolution and then applying 420 chroma subsampling allows to overcome most of this.
Yeah, that's kinda what I meant - I won't have any intermediate steps, it'll dump in 4K straight away.
feos wrote:
BTW, since Tails is so critically important in this run, won't it be better to keep him non-transparent at all? I guess he's half-transparent when off-screen, but this is one of the biggest features here, so making it more obvious would help. I doubt not knowing when he is off-screen will make anyone angry.
The fact that he goes semiopaque is because every object is semiopaque while not being rendered; I feel like adding more special cases is gonna make the code look pretty terrible. Hell, making it semiopaque to begin with already makes it rather complex, but it is a bit important because that lets you easily see why you can pass through objects at that particular time. I might switch it to being e.g. desaturated if it's deemed too distracting. Oh, and about the NUT encoding: I'll definitely try that, just in case image2pipe has some weird perf issues.
Posted: 6/27/2019 5:59 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
feos wrote:
https://vadosnaprimer.github.io/feos.vs.youtube.2160p.html
Thank you! This will be very useful.
feos wrote:
Do you render at 1x scale? It's quite possible that rendering at 2x scale could allow you to preserve much higher quality in general while increasing the ratefactor significantly to reduce the size. It's worth using a short test clip to see what youtube does to your video, be it 4K or 8K.
I'll render the final output in 4K in order to remove the 4:2:0 chroma subsampling. 8K would probably be overkill, unless someone else does it for me (prepare to spend 200GB on temp files!)
Posted: 6/27/2019 5:13 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
feos wrote:
Does it keep all the intermediate images too? I'm seeing Rust for the first time :D
Yeah, sorry for not keeping the codebase commented - I might have to do that for most of it so that others could learn and contribute. The intermediate images are stored in a hashmap since they could arrive out of order, since they are rendered in a thread pool. They are then removed from the hashmap when there's enough to go in order. Rust has move semantics, meaning that when you have ownership of some data, you are the sole owner, and it is guaranteed that dropping the value by moving it out of scope will deallocate the buffer, unlike in e.g. GC'd languages.
Posted: 6/27/2019 4:21 PM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
feos wrote:
Congrats on finishing it!!!
Well, it's nowhere near finished - it's definitely just the draft and I'll need to polish it a lot more.
feos wrote:
Please tell me which code snipped dumps the framebuffer currently. There are lots of ways to reduce the filesize without killing the quality.
https://hyper.is-a.cat/gogs/x10A94/atlas-s3k/src/master/src/main.rs#L244 As of right now, this is the code that accepts the images being encoded in a thread pool and uses a BMP encoder for the images, which pipes the output into ffmpeg. Actually, now that I think of it, I could probably squeeze out some more performance by using an RgbImage as the target, since then the encoder would not have to do much to actually encode it.
Posted: 6/27/2019 5:47 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
WST wrote:
Just curious, how much space do you need?
Now that I think about it, I'm pretty sure I'll manage. I'm currently using the veryfast preset to improve framerate, but if I wanna make it final I might as well use veryslow, which would improve compression significantly. I think the biggest bottleneck is just piping the data - I think it would be more efficient to use ffmpeg directly as a library, but sadly I have absolutely no idea how to do so. If someone has any sort of idea on how to feed the ffmpeg library a bunch raw image buffers and get a video out of it, I'm all ears, the project is open-source: https://hyper.is-a.cat/gogs/x10A94/atlas-s3k/
Posted: 6/27/2019 1:02 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
Link to video I got the encode to the point where it's possible to render most of the game with negligible artifacts. At some point in the future I'm gonna have to make it so that it's uploaded to youtube while it's encoded into 4k, because I simply do not have the space. Still not sure how to handle backgrounds. For now it's a checkerboard. Might be a bit jarring, and I might make it bigger.
Posted: 3/31/2019 10:20 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
Since the publishing of the movie is fairly soon, I decided to work some more on the Atlas encode. This time with a dynamic camera that needs a bunch of fixes to make transitions not as jarring. Link to video
Posted: 1/20/2019 2:26 AM
(Edited: 6/4/2022 5:10 PM)
Post subject: Marble Marcher
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
I've tried to run Marble Marcher, the new game that uses fractal geometry for level layout, but it looks like it doesn't quite replay inputs correctly, and after 2-3 frames of playback it starts using real inputs instead of emulated ones. The game's open-source, so it'll likely be much easier for you to understand what's going wrong. TAS runs of this game would likely be extremely impressive due to how precise the collision in this game is. Link to the game: https://github.com/HackerPoet/MarbleMarcher
Posted: 1/4/2019 11:23 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
r57shell wrote:
my main goal was to be able play the game
Yeah, your mod is great for realtime playthroughs, but for a TAS like this where the camera must be dynamic to make sure that you actually see where both characters are on the level, it should account for future frames as well, which is not very possible when running a realtime renderer. It's also pretty crucial to be able to hardcode things at certain points in the run, since it's unlikely that I'd be able to nail the presentation for everything in the entire game.
Posted: 1/4/2019 10:58 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
r57shell wrote:
IDK what you doing, but I'm planning to add encoding method into my sonic atlas... Just because of this event. Here is links: http://tasvideos.org/forum/viewtopic.php?p=470260#470260
Here's my code: https://hyper.is-a.cat/gogs/x10A94/atlas-s3k/src/master I did not want to keep this specific to Gens, considering there's plenty of Bizhawk runs out there, or specific to Genesis, even. Thus, it'll be possible for me to create Atlas encodes of other games, even if they're slightly less polished than those done with video editing. As such, the most common denominator that I could get to work on Linux was.. creating a savestate for each frame, plus some mid-frame data. This isn't very optimized, but it really doesn't need to be, given that it creates a 1080p video at around 15fps on my machine, which is fast enough for debugging. If you somehow get it to work on windows, I'll put it on github so you can make a pull request.
Posted: 1/4/2019 5:49 AM
Quote
Selicre
He/Him
Experienced Forum User
Joined: 1/4/2019
Posts: 17
Location: RU
Evil_3D wrote:
But a friend is working on a complete Atlans encode for this run, would take a while but will worth it
Speaking of which, here's HCZ2: https://www.youtube.com/watch?v=sxzHItPChgQ It's still not perfect, and I still have yet to find out how to hide the HUD more efficiently as well as draw the plane B data, but it does show how Tails is used offcamera in most places. Moving forward, I'm going to change it to focus on the interpolated midpoint of Sonic and Tails, so that way Tails doesn't end up off-camera.