Not that I haven't tried already - I know about frame blending, interpolation, motion blur etc. it's just that I havent had any success with it. I'll try to keep it as short as possible.
I have the following tools at my disposal: Avisynth + SVPflow + MVTools,After Effects CS5 + Twixtor. Also installed Sony Vegas but I havent used it yet.
Youtube completely rapes every 60fps footage I upload - its incredibly choppy, too choppy.
I know that it converts everything to 30 but their algorithm is so bad, its unbelievable.
I converted my 60fps sample to 30fps by just removing half of the frames (ChangeFPS) and when I uploaded it to youtube it was noticeably less choppy than the 60fps footage that had been converted by youtube's own algorithms!
Now I know people like this guy: youtube.com/watch?&v=t6e1N2Bot4k have managed to convert their 60 footage to 30 using interpolation and motion blur vectors, and the results are exactly what I'm looking for.
I haven't had any luck with After Effects, I interpreted the footage as 30, sped up the footage by 100% and added motion blur - I did not notice any difference. Also no luck with Twixtor.
I used the default avisynth frameblending and the results were horrible: it was just as choppy as the regular 30fps footage with half the frames and on top of that there was really bad ghosting.
I tried writing a SVPflow script for Avisynth (http://www.svp-team.com/wiki/Plugins:_SVPflow) but I don't understand the syntax and its too complicated.
Apparently I'm a complete dumbass and I've just grown tired now of trying to accquire the necessary info all by myself.
I either need a step by step dummy instruction for one of those expensive video editing programs people are supposed to pay money for, or a finished avisynth script using the svp/mvtools plugins that I can just copy paste and use in MeGUI.
converting 60fps to 30fps is as simple as removing half frames. now if you needed to convert 60fps to 24fps that's a whole different story.
i don't know why you keep talking about motion blurs and what not, that's not needed, you are just messing up your footage. 30fps will never look as smooth as 60fps.
Motionblur in AE does not do what you think it does.
It's only for motion of animated layers.
What you want is frame blending.
You can enable vector motion from there.
But 60 is a direct multiply of 2 from 30 so the absolute best quality is to just drop every other frame.
Probably best performance through avisynth.
Use vector motion for stuff where it's not a direct integer multiplier like that.
converting 60fps to 30fps is as simple as removing half frames. now if you needed to convert 60fps to 24fps that's a whole different story.
i don't know why you keep talking about motion blurs and what not, that's not needed, you are just messing up your footage. 30fps will never look as smooth as 60fps.
That's bad. Better is to blend two frames together. This is what is called "motion blur".
sigh. best, cleanest, most accurate way to convert 60fps vid to 30fps is to remove every second frame, that's all there is to it. you won't get smoother result than this.
frame blending is necessary only when converting to something like 60fps to 24fp. but still even with frame blending such conversions are not recommended, as any video professional would tell you. there are expensive plugins that try to rectify this somewhat (Magic Bullet Frames) but this is another topic and OP doesn't need any of this.
That's incorrect. When a camera captures 30 fps directly, a "blur" effect is created. When you capture at 60fps, the "blur" effect is much less pronounced, because you have double the temporal resolution. If you cut every second frame, you lose half of the information. If you blend, you retain the information as much as possible. Similar to how you resize an image to 50% of its original size. If you do a simple pixel remove, you lose 3/4 of the information. That's why you do various of algorithms that blend pixels to achieve best result (as much as possible).
The difference between 60fps->30fps and 60fps->24fps is that the former is a rather simple blend, while the other is nonlinear and is more complex.
Well the best result is the slowest one. Activate the best motion blur setting in after effects and set the video to 30fps.
It should take quite some time to render but will look as good as possible. Leaving game footage at 60fps without motion blur will let it look too crisp like sports recordings. If you take real footage with motion blur, dropping every second frame will lead to loss of information and may degrade video quality.
Yeah thanks, I rendered some versions of a sample section now, using various motion blur/frame blending settings in AE. It really gives an illusion of 60fps, at least when played back locally.
Is there any specific setting you can recommend that looks best for video game footage?
AE itself has two kinds of frame blending (regular and pixel-motion), then there are more than one different motion blurs on top of that, each with shutter settings and sample counts.
I also have the Twixtor plugin, which comes with its own versions of those algorithms.
Its hard to find the middle ground between too blurry and not smooth enough.
Rendering takes ages, 15 minutes for a 15 second sample, is there any way to accelerate this by getting help from the GPU? The total footage I'd need to transform is about 80 minutes (!), under the assumption that AE won't slow down rendering bigger files this would take me around 5 to 7 days to render. INSANE!
I don't even know if I will go through with this, or just quickly encode a version with half the fps. Its only made worse by the fact that ALL youtube videos are choppy for me, even on different browsers, with and without hardware accel. The frame blend samples I uploaded look smoother, yes, but there are still bouts of choppy during playback. I never have those bouts locally! Footage and trailers that is running ultra smooth at 30fps in MPC and VLC is choppy as hell on youtube. Either there is sum ting wong on my side or youtube and flash are just a huge pile of shit.
Black, white, black, white ... 60 times per second would look somewhat like gray to you, depending on the sensitivity of your eyes. But your example is a bad one. Take an image with the alternating black white pattern and scale it down and you are left with a shit result that is either gray in best case or has a moire effect in worse case. And still, when you resize an image in Photoshop, you select "Best" and it chooses bicubic for you and blends the pixels. In extreme examples, like yours, it doesn't work well. In normal cases, it works better than other solutions (like taking every other frame).
There are more advanced solutions, where motion vectors are first discovered, and then blending is performed on per-vector basis. These can work well, but in my experience, they almost always introduce some artifacting, and the artefacts introduced are not worth the extra smoothness in most video types.
We need something like a "forced reverse YADIF" working on non-interlaced input, with motion taken into consideration, so parts of the frame/image which dont change (background) are not blended, and moving parts are blended variably, according to speed & size.
that's already done
also this thread is full of derp. blending isn't needed when converting 60->30 like i said 3 times already or so. but if you want "fake" smoothness with artifacts good luck with frame blending. not sure what derpatan is going on and on about, i guess he got a case of "i'm afraid to be wrong" sickness. there is a reason frame blending is off by default in After Effects, because it's not natural and is mostly used along with time stretching and time remapping, not a simple 60fps->30fps conversion.
Last edited by rgb#000 on Wed, 17th Jul 2013 18:32; edited 1 time in total
That video is a prime example of how horrible the "intelligent" frame blending is. Take a look at all the artifacts in leaves, around the arms and between the legs.
But how is that video appropriate here? One is about compressing data while the other is about generation of data. Data compression is a lot easier and errors are less apparent than having to generate more data from less data.
well i guess i said all i wanted to say in this thread. i'm not even sure what was OP's problem at this point, he wants to magically make 30fps look like 60fps using various kinds of blurs and what not. so stupid. that's like console devs using blurs to hide crappy textures
That video is a prime example of how horrible the "intelligent" frame blending is. Take a look at all the artifacts in leaves, around the arms and between the legs.
But how is that video appropriate here? One is about compressing data while the other is about generation of data. Data compression is a lot easier and errors are less apparent than having to generate more data from less data.
Generation of data is actually the easier one now.
Some artifacts are unavoidable but the algorithms for adding additional info to slowed-down footage or multiplying the framerate of videos are pretty far advanced now.
The SVP plugins can transform any video to 60fps in realtime (while watching) by adding and guessing frames and the result is 95% flawless. I only use it for nature documentaries though (Attenborough's Micro Monsters). Tried it in Dexter, Falling Skies, Luther too but I just can't get used to it and keep focusing on whats going on in the background because everything is so fluid suddenly.
But these functionalities are the exact opposite of what I wanted to achieve, because they actually add frames.
RGB is right, all the frame blend trickery and motion blur is a crutch at best and due to the time it takes them to render its just not worth the hassle.
I managed to encode a sample in After Effects with pixel-motion blending and cc motion blur which truly focused on the moving parts and kept the background (mostly) intact but its just not good enough.
It ends up like a mutated monster, thats slightly smoother than regular 30fps but can't hold a candle against "true" 60fps (that "speeding-up" effect just isnt there)
I believe I mixed up the two blend modes earlier...
Pixel-motion is only for slowing down/ super slo mo, in case of a 60 fps video converted to 30 it starts to work only once you slow it down by more than 100%.
Its actually adding missing frames, guessing where the entities would be at next.
You can slow down Tekken by 1000% and let pixel-motion do its magic but at normal speed it does nothing.
Maybe you can slow down the footage by factor 10, let pixelmo add frames in a first pass, and then in a second pass use the footage with the additional frames and speed it up by a factor of 10 again. Or is this just moving in circles?
It will probably remove most frames again and you'll get a new choppy hybrid with some original frames and some of the surviving guess-frames?!
Best I got is just using regular blend and optionally some motion blur on top of it to lessen the ghosting. This smoothens the video a bit, at the cost of image clarity.
But maybe you have more luck..looking forward to your comparison.
I am playing with the sample. First problem I see is the video is not 60fps precisely, but 59.94 (NTSC). Converting directly to 30fps is not as straightforward anymore (unless you want to convert to 29.97 NTSC). What I am doing is assume it's 60. I will do my test without audio.
Signature/Avatar nuking: none (can be changed in your profile)
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
One is about compressing data while the other is about generation of data. Data compression is a lot easier and errors are less apparent than having to generate more data from less data.
