waifu2x

[tormento]

Can we have a port of this?

https://github.com/nagadomi/waifu2x

[Reel.Deel]

Can we have a port of this?

https://github.com/nagadomi/waifu2x

There's already an AviSynth port (rather slow): https://github.com/sunnyone/Waifu2xAvisynth

Although not as extensive as VapourSynth's selection:

https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Waifu2x-caffe
https://github.com/HomeOfVapourSynthEvolution/VapourSynth-waifu2x-opt
https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Waifu2x-w2xc
https://github.com/Nlzy/vapoursynth-waifu2x-ncnn-vulkan

[tormento]

There's already an AviSynth port (rather slow)
It doesn't get advantage of GPU, I bet it's slow. :D

[Sparktank]

It doesn't get advantage of GPU, I bet it's slow. :D

I tried the AVS version once, and it crashed instantly.
Learn VS for this one.

[ChaosKing]

Or try http://avisynth.nl/index.php/VapourSource

[real.finder]

if someone plan to port any, it's better to chose vulkan one, cuda is Monopoly to nvidia

[real.finder]

I tried the AVS version once, and it crashed instantly.
Learn VS for this one.

it was x86 only no x64, so indeed you get RAM limit crash

[Sparktank]

it was x86 only no x64, so indeed you get RAM limit crash

Now, I remember. Wow, that really takes me back when I used to use x86.

I'll give VapourSource a try.

[tormento]

if someone plan to port any, it's better to chose vulkan one, cuda is Monopoly to nvidia
Or, better, OpenCL.

[ChaosKing]

Vulkan is the future!

[real.finder]

better to make it like this



gpu: Controls the environment to use.

0 = disable GPU
1 = auto detect. It will run on the first available environment in the following order:
CUDA
Vulkan
AMD OpenCL
FMA
AVX
Intel OpenCL
SSE3
OpenCV filter2D
2 = force to use OpenCL

in https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Waifu2x-w2xc

OpenCL still has features since it run even on CPU only

[ReinerSchweinlin]

Lookup hybrid by Selur, he has Waifu2x already implemented (filter-vapoursynth-line...) Usage: Resize by factor 1,2 or 4 (1 for "cleanups"), then select waifu2x as the resizer in the mentioned "Line" tab..

[FranceBB]

There's already an AviSynth port (rather slow): https://github.com/sunnyone/Waifu2xAvisynth


Rather slow? I wanted to make a screenshot comparison of a 720x480 NTSC DVD I ripped a while ago and I had on my HDD and it took 18 fucking minutes on my crappy i7 4c/8th 2.60GHz for one single frame.
In comparison, QTGMC is blazing fast xD

It doesn't get advantage of GPU, I bet it's slow. :D

CPU only is a nightmare; 18 minutes for 1 single frame to go from 720x480 anamorphic to 1440x960 anamorphic on a 4c/8th. It can't be used...
GPU acceleration is a must.

I tried the AVS version once, and it crashed instantly.

Well it didn't crash but it used 2 GB of RAM for itself and took a hell lot of time for 480p to 960p; I bet it will crash if I try higher resolutions...
(EDIT: it didn't crash for higher resolution)


Besides, it only supports y8, yv12, yv16, yv24 which means it's stuck on 8bit planar.


Anyway, lo and behold, this is the single anamorphic yv12 frame that took 18 minutes to encode:

Original:


DGDecode_MPEG2Source("\\VBOXSVR\Share_Windows_Linux\Production\RAW\417-423\421.d2v")


tfm(mode=1,pp=5,slow=2,micmatching=2)
tdecimate()



Waifu2x:

Waifu2x(nr=0, scale=2, models="C:\Programmi\AviSynth\models")

NNEDI3:

nnedi3_rpow2(cshift="Spline64ResizeMT", rfactor=2, fwidth=1440, fheight=960, nsize=4, nns=4, qual=1, etype=0, pscrn=2,
csresize=true, mpeg2=true, threads_rs=0, logicalCores_rs=true, MaxPhysCore_rs=true, SetAffinity_rs=false, opt=3)

Bilinear:

BilinearResizeMT(1440, 960)

Bicubic:

BicubicResizeMT(1440, 960)

Lanczos:

LanczosResizeMT(1440, 960)

Spline64:

Spline64ResizeMT(1440, 960)


Screenshots:

Original:
https://i.imgur.com/lksozYG.png
NNEDI3:
https://i.imgur.com/HBy6GhE.png
Waifu2x:
https://i.imgur.com/hm1DTt9.png
Bilinear:
https://i.imgur.com/sBql880.png
Bicubic:
https://i.imgur.com/qBzMo97.png
Lanczos:
https://i.imgur.com/iQHV7ET.png
Spline64:
https://i.imgur.com/vAFETNr.png
Point:
https://i.imgur.com/yZ3gHfA.png

Link to Imgur: Link (https://imgur.com/a/3sQkfH0)

Clearly Point is the worse.
My chart is (from best to worse):

NNEDI3 (with Spline Kernel)
Spline64Resize
LanczosResize
BicubicResize
BilinearResize
PointResize

NNEDI3: it's alright I guess.
Spline64: very sharp, but there's a bit of ringing around the lobe of the ear.
Lanczos: still sharp, but there's still ringing.
Bicubic: a bit blurry, ringing still there but less because of the blurriness.
Bilinear: bluerrier (is it even a word?) than Bicubic.
Point: total crap, as expected by nearest neighbor.

All the test completed in a matter of seconds; the only one who took 18 fucking minutes was Waifu2x and it didn't even use my CPU at 100%, it was sitting at 12% (with waifu2x internal denoise disabled) and 25% (with waifu2x internal denoise enabled). You know, after waiting 18 minutes staring at the task manager allocating RAM for 1 single frame I expected the upscaled image to be so much better than the one made by its competitor, but if you compare Waifu with NNEDI 8bit it doesn't really add anything to it other than making eyes a bit more white and make the forehead protector color look a little bit different... It still has plenty of ringing as well...
In other words, I would totally go for NNEDI with Spline as kernel instead of bothering with this waifu thing for three reasons: 1) NNEDI is way faster and gives similar results 2) it can work in 8bit or 16bit while waifu2x is limited to 8bit 3) it can be used for everything, not just anime


EDIT: For the records, I tried to go up to x8 and it took 1h and 47 minutes for 1 frame on my 4c/8th using it at 18%...

I mean, guys, is it really worth it...? I don't think so...

[poisondeathray]

All the test completed in a matter of seconds; the only one who took 18 fucking minutes was Waifu2x and it didn't even use my CPU at 100%, it was sitting at 12% (with waifu2x internal denoise disabled) and 25% (with waifu2x internal denoise enabled). You know, after waiting 18 minutes staring at the task manager allocating RAM for 1 single frame I expected the upscaled image to be so much better than the one made by its competitor, but if you compare Waifu with NNEDI 8bit it doesn't really add anything to it other than making eyes a bit more white and make the forehead protector color look a little bit different... It still has plenty of ringing as well...
In other words, I would totally go for NNEDI with Spline as kernel instead of bothering with this waifu thing for three reasons: 1) NNEDI is way faster and gives similar results 2) it can work in 8bit or 16bit while waifu2x is limited to 8bit 3) it can be used for everything, not just anime


EDIT: For the records, I tried to go up to x8 and it took 1h and 47 minutes for 1 frame on my 4c/8th using it at 18%...

I mean, guys, is it really worth it...? I don't think so...

Agreed - not worth it for avs... But that's the old avs version. It's very slow (no optimizations), and has limited settings, uses older model - it was only a proof of concept. It looks nothing like modern implementations...

You need to use vapoursynth-caffe version, takes about 1 sec (depending on GPU), more settings, different models, works in 32bit, cleaner, better lines and less ringing. That's why people are asking for an avs port...
https://postimg.cc/KKbNwSnz

[real.finder]

I just note there are another port of waifu2x optimized for video called dandere2x https://github.com/aka-katto/dandere2x

[tormento]

GPU acceleration is a must.
That's what I would like to see in the close future, for every plugin that could benefit from GPU architecture.

I proprosed OpenCL because it's the easiest to program and port to CPU also, becoming "transparent" to users.

[tormento]

better to make it like thisgpu: Controls the environment to use.
Agree but it would mean lot of work for programmers.

To me even OpenCL only is more than enough.

[FranceBB]

Agreed - not worth it for avs... But that's the old avs version. It's very slow (no optimizations), and has limited settings, uses older model - it was only a proof of concept. It looks nothing like modern implementations...

You need to use vapoursynth-caffe version, takes about 1 sec (depending on GPU), more settings, different models, works in 32bit, cleaner, better lines and less ringing. That's why people are asking for an avs port...
https://postimg.cc/KKbNwSnz

I see...
Yes, the new implementation looks way better than the old implementation, it's sharp and it definitely has less ringing.
Now I understand why people want a modern version ported to AVS...

That's what I would like to see in the close future, for every plugin that could benefit from GPU architecture.

Yes, that would definitely be welcomed. At work we have servers with NVIDIA Quadro GPUs that are currently not used other than for x264 OpenCL accelerated encoding, which is a total waste, honestly. I would love to see plugins supporting both CPU and GPU acceleration.


To me even OpenCL only is more than enough.

Uhm... I would rather have a single plugin that can do both CPU and GPU and for which you can decide which one is gonna use or let it do automatically for you. This is to avoid having two separate version of plugins to maintain. Like, let's suppose you're in a world where everything is hardware accelerated with GPUs: you would have things like dfttest and dfttestGPU for instance and you would have to maintain both... it would be a nightmare for developers.
I like real.finder's idea; after all he's a *real* finder of ideas XD.

better to make it like this



gpu: Controls the environment to use.

0 = disable GPU
1 = auto detect. It will run on the first available environment in the following order:
CUDA
Vulkan
AMD OpenCL
FMA
AVX
Intel OpenCL
SSE3
OpenCV filter2D
2 = force to use OpenCL



That would be very good but it would be very hard and time-consuming to develop. Even something like having just the ability to either choose CPU only or GPU-acceleration, where the GPU acceleration can be whatever the developer chooses like CUDA or OpenCL etc
Let's be real, no one is going to re-implement a filter (that has already been written) with different GPU implementations.
I know that OpenCL can be used by GPUs from both NVIDIA and AMD and that it's impossible for AMD/ATI users to use CUDA plugins, however I have NVIDIA Quadro GPUs at work and NVIDIA RTX at home, so for me CUDA is fine; heck, it's even better than OpenCL in my case 'cause OpenCL runs slower on NVIDIA, so I'm happy with that (for instance the plugins developed by Donald Graft run just fine). On the other hand, I do understand why AMD/ATI users complain...

[MeteorRain]

I don't think opencl on CPU is well supported. No?

[feisty2]

there're much more powerful neural net based models than waifu2x and nnedi3 (https://github.com/kice/vs_mxnet), take a look at ESRGAN (https://github.com/WolframRhodium/Super-Resolution-Zoo/tree/master/ESRGAN) and RCAN (https://github.com/WolframRhodium/Super-Resolution-Zoo/tree/master/RCAN)

[ReinerSchweinlin]

Yes, Waifu2x is fine for anime and stuff - but in no way potent for real-life stuff..

As mentioned above, take a look at "hybrid" from selur - he already has a working implementation of the vapoursynth ports with vulkan etc... No need to re-invent the wheel for avisynth with CPU-based OPEN CL ....

[tormento]

I don't think opencl on CPU is well supported. No?
Yes it is on integrated GPU and I think it's easily portable from existing C++ code with appropriate tools.

[tormento]

Yes, that would definitely be welcomed. At work we have servers with NVIDIA Quadro GPUs that are currently not used other than for x264 OpenCL accelerated encoding, which is a total waste, honestly. I would love to see plugins supporting both CPU and GPU acceleration.
Whenever your company will decide to buy next generation and dump current, please send me a PM, seriously. :cool:
This is to avoid having two separate version of plugins to maintain.
AFAIK, at least with CUDA and perhaps OpenCL it's really easy to port from C++ to GPU accelation. We should ask Khanattila how did he port NLMeans to KNLMeansCL.
so for me CUDA is fine; heck, it's even better than OpenCL in my case 'cause OpenCL runs slower on NVIDI
CUDA has many many tools and resources to help programmers. AMD has great hardware but really poor support and drivers.

[feisty2]

NTSC master
https://i.imgur.com/jFqz13W.png
https://i.imgur.com/I0lYFRr.png

scaled to 2K by ESRGAN
https://i.imgur.com/U6EnIFQ.png
https://i.imgur.com/h4tY6rw.png

[feisty2]

ur welcome to challenge the results with waifu2x, nnedi3, traditional resizers, anything you like, and see if the quality could be nearly as good

[feisty2]

note that ESRGAN is already available to vaporsynth thru vs_mxnet, you can turn your SD masters to native 2k anytime you like, the quality is levels above waifu2x or nnedi3, I therefore find all this waifu2x discussion totally meaningless

[poisondeathray]

note that ESRGAN is already available to vaporsynth thru vs_mxnet, you can turn your SD masters to native 2k anytime you like, the quality is levels above waifu2x or nnedi3, I therefore find all this waifu2x discussion totally meaningless

ESRGAN is also available through vsgan
https://github.com/rlaPHOENiX/VSGAN

But that image looks nothing like a "native 2K image". It's a huge overstatement



But waifu2x is meant for anime sources, and was trained on anime sources - how is it "meaningless ? "

Or, do you have an ESRGAN model that gives similar or better results than waifu2x on typical anime sources ?

It's all about the training used and the specific model. I tried some public distributed anime trained models for ESRGAN , and the results are significantly worse

[feisty2]

the problem is not about anime or photos, waifu2x is a really "early" project, its models are too small and do not have enough capacity compared to state-of-the-art models in the academia. if someone's gonna port neural net based resizers to avisynth, I think it makes more sense to directly go for the modern models.

[feisty2]

But that image looks nothing like a "native 2K image". It's a huge overstatement



still, small models like waifu2x and nnedi3 can never get close, I have posted the original SD images, try for yourself.



I tried some public distributed anime trained models for ESRGAN , and the results are significantly worse



I am 99.9% sure that, given the same training set, anime or not, ESRGAN will produce much better results

[poisondeathray]

the problem is not about anime or photos, waifu2x is a really "early" project, its models are too small and do not have enough capacity compared to state-of-the-art models in the academia. if someone's gonna port neural net based resizers to avisynth, I think it makes more sense to directly go for the modern models.

I agree

still, small models like waifu2x and nnedi3 can never get close, I have posted the original SD images, try for yourself.


Of course, it's just stupid to even try waifu2x on non anime source , and nnedi3 is known for coarse lines, very old neural network (but Tritical was years ahead of almost anyone)




I am 99.9% sure that, given the same training set, anime or not, ESRGAN will produce much better results

Probably, but it takes lot of time and resources to train.

Waifu2x has many different GUI's for it, several types of pretrained models available, some with lua scripting in other programs (aviutl) , many different implementations supporting AMD, Nvidia) - it just seems so mature and that it should be easier to port to avisynth.

Many of the newer research projects are python based, pytorch, etc...so they should be able to be work in vapoursynth much more easily, than something like avisynth

[feisty2]

Of course, it's just stupid to even try waifu2x on non anime source

there're waifu2x models optimized for photos iirc, have a try and compare the results with ESRGAN if you can find them

edit: here (https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Waifu2x-w2xc/tree/master/Waifu2x-w2xc/models/photo)

[poisondeathray]

there're waifu2x models optimized for photos iirc, have a try and compare the results with ESRGAN if you can find them

edit: here (https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Waifu2x-w2xc/tree/master/Waifu2x-w2xc/models/photo)

I tested the waifu2x photo models before about a year ago - I can tell you now, it won't even be close. The waifu2x "strength" is only typical anime content.

Which ESRGAN model did you use above?

[feisty2]

the official one (https://github.com/WolframRhodium/Super-Resolution-Zoo/tree/master/ESRGAN)

also the results above pass for 2k to me, it surely ain't high quality 2k, it looks a bit like the quality of 16mm film. native 2k could be u know, slightly out of focus and stuff and not necessarily very sharp or with extremely delicate details.

screenshot took directly from a 2k master (filmed in 2018), does it also look "fake" to you?
https://i.imgur.com/fi7Wkz2.png

[zorr]

As good as ESRGAN is, it's still meant for still images. When applied to video it may not be temporally coherent. Has anyone thought about porting GANs which are meant for video? Such as TecoGAN (https://github.com/thunil/TecoGAN) (and I'm sure there are many more).

[poisondeathray]

also the results above pass for 2k to me, it surely ain't high quality 2k

Not for me; it screams "upscale" to me. It might be "less bad" than other algorithms, but there is 0% chance anybody that works with video or image would mistake that as native 2K.


screenshot took directly from a 2k master (filmed in 2018), does it also look "fake" to you?


Not "fake" per say, but that shot has different characteristics with the motion and shutter blur . It's also framed differently at a distance

There is no way if you had a comparable shot up close that you would have the coarse detail characteristics, image artifacts like that Britney upscale

[feisty2]

There is no way if you had a comparable shot up close that you would have the coarse detail characteristics, image artifacts like that Britney upscale

close-up shots can have coarse detail characteristics, I don't wanna pollute this thread with any more huge screenshots, gonna use a link instead. (https://i.imgur.com/kLPbOn8.png)


Not for me; it screams "upscale" to me. It might be "less bad" than other algorithms, but there is 0% chance anybody that works with video or image would mistake that as native 2K.

have you tried scaling the original image with, say, spline64? that's what upscale normally looks like, then there're nnedi3 and waifu2x which produce sharper edges but the result becomes "oil-painted", neither looks like the result of ESRGAN. The "eye-lash" or the whole "eye" region looks very native hi-res to me.

[poisondeathray]

close-up shots can have coarse detail characteristics, I don't wanna pollute this thread with any more huge screenshots, gonna use a link instead. (https://i.imgur.com/kLPbOn8.png)
Yes, but that looks like native 2K. There are fine details, like fine grain, lace patterns in the eye patch. You can see strands of hair clearly, not edge clumps that look upscaled. It fits together and coherently - There is nothing "off" like compression artifacts, ringing that are dead giveaways in the Britney upscale. Brit has some parts that are upscaled nicely , but others that are not - it's a hodgepodge, that' s why your brain (or my brain) screams upscale

Nobody would think this was an upscale, almost everyone (that has any experience) would think Britney was. You might be able to pass Britney off to the uninitiated or lay person, not to anyone that works in image or video industry.




have you tried scaling the original image with, say, spline64? that's what upscale normally looks like, then there're nnedi3 and waifu2x which produce sharper edges but the result becomes "oil-painted", neither looks like the result of ESRGAN. The "eye-lash" or the whole "eye" region looks very native hi-res to me.

Of course, but they all look non native 2K.

ESRGAN is "less bad", and it does a generally better job with finer details. But there are telltale signs that make it look like an upscale and not native 2K.

If you used the original models very much, certain texture patterns are prone to serious errors as well . Certain frames are unusable because of this

The flicker or temporal inconsistency was mentioned earlier too; it's not that great on video because of those issues

[feisty2]

I'm not sure what you mean by "edge clumps", I found a native HD close-up shot of britney in another video (https://i.imgur.com/m4Npq27.png) and the hair also looks "clumpy", probably just the texture of her hair.

all these were shot in circa 2010, when music video production began shifting towards 1080p. a lot of HD videos shot around that time or earlier have this "coarse" look, you can tell it's HD, but it's not high quality HD, it definitely does not squeeze out the full potential of 2k resolution. you just know the image could still be much "finer" and more "delicate" at 2k. the gap between SD and HD wasn't all that huge since SD videos at that time usually look much "finer" at their native resolution.

more recent shots have managed to get rid of this "coarse" look, they look more "HD" than them shots in 2010, the image looks much more "delicate" and you can tell instantly that it has exploited every last bit of 2k resolution potential. (https://i.imgur.com/QJWrQQG.png) this is the "real" HD quality I'm talking about. I'm not quite sure what changed in the production process of recent years that led to such quality boost.

[FranceBB]

I'm not quite sure what changed in the production process of recent years that led to such quality boost.

Lens, cameras, high bit depth used all the time (from recording to post-processing) and logaritmic curves to keep as many details as possible in terms of light captured (more stops) even if the final file was then going to be graded down to linear BT709 100 nits.
Not that cameras of the past were bad, but we made huge step forwards compared to years ago as technology became so much better. And this was all years before HDR was even a thing! You can see how cameras became much more advanced to record frequencies than what the official standardized specs were allowing broadcasters to air... We squeezed everything we could for years out of the FULL HD standard...

[feisty2]

I think it probably has more to do with lens and cameras. older cameras can record 1080p videos but that's highest resolution they can handle, while modern cameras can record 6k or 8k videos and when downscaled to 2k, it should look much "finer" than footage shot by 2k cameras.

[poisondeathray]

I'm not sure what you mean by "edge clumps", I found a native HD close-up shot of britney in another video (https://i.imgur.com/m4Npq27.png) and the hair also looks "clumpy", probably just the texture of her hair.


Clumps can be a hairstyle; but you would expect more visible separation if it was real HD, there are areas that look like blocks instead of being composed of hair strands.


I think it probably has more to do with lens and cameras. older cameras can record 1080p videos but that's highest resolution they can handle, while modern cameras can record 6k or 8k videos and when downscaled to 2k, it should look much "finer" than footage shot by 2k cameras.

That is a large part of if you compare with modern cameras; It's oversampled when you downscale to HD . Even cheapo consumer cameras today can have higher actual resolution (as measured by zone plates, charts) when downscaled, than expensive HD cameras back then

The image does not look like authentic HD. Yes, it's shot with shallow DoF, partial high key, but you would expect finer hair ,brow , and skin details even with thick makeup. It upscales some vellus hair on the face, it's semi-convincing in the 1st , but poor in the 2nd. But the artifacts ruin the illusion. Parts of it are upscaled ok, but parts are not - that inconsistency also ruins the illusion

[feisty2]

a markov discriminator (https://arxiv.org/pdf/1604.04382.pdf) could probably help with the inconsistency problem. I don't currently have access to GPU servers in my lab so I cannot verify if the idea works.

still, the ESRGAN results are greatly superior than small models like waifu2x, at least certain parts of the upscaled image look native, while other parts look "passable". I've been dealing with generative models for so long and I know model size is a key factor to the performance of generative models, it's all about memorizing the training set, but in a smart way. small models simply do not have the capacity to memorize the distribution manifold of natural images. people should give up the idea of doing upscales with small models.

[FranceBB]

I think it probably has more to do with lens and cameras. older cameras can record 1080p videos but that's highest resolution they can handle, while modern cameras can record 6k or 8k videos and when downscaled to 2k, it should look much "finer" than footage shot by 2k cameras.

Oh, yes, I was talking about the 2009-2013 FULL HD era to be fair, but definitely new modern cameras from 2018-2020 can record at a higher resolution and the downscaled version will definitely look very sharp.

It's an unfair comparison, but these are FULL HD footages that were originally shot in 8K by Jacob Schwartz with a Red Monstro: Link (https://imgur.com/a/KloBNlg)

Anyway, I like the results you posted made by ESRGAN.
Of course it will never be like a real HD / FULL HD footage but for an upscale it looks pretty good to me. If it can be made coherent and avoid temporal inconsistency, then it's gonna be fine for me.

[ReinerSchweinlin]

Topaz Labs is just working on an AI Upscaler for MAC and WIN. a Demo can be downloaded. I am doing some stuff betatesting for them - the last few weeks the frameworks have been updatet and refined, now more training on the models is to be done. Still an early stage, but if you guys want to play around a little, look it up:

https://topazlabs.com/video-enhance-ai/

[real.finder]

I think this thread is getting off topic and ad posts!

this thread about waifu2x and it's forks like dandere2x

ESRGAN was already has https://forum.doom9.org/showthread.php?p=1886029#post1886029

[ReinerSchweinlin]

Well - Dandere is no "fork", but a GUI which adds a speedup function to make the upscale faster by determinig parts of frames which don`t change and therefore can be simply copied from frame to frame (old style anime is good for that in some cases). Video2x is another GUI for waifu2x using the "picture by picture" approach, using a lot of space for intermediate PNG Files.
Hybrid is a GUI which already has two vapoursynth variants implemented which are piped into the workflow, so you simply can select waifu2x in Cuda and Vulkan variants as scalers... There are other GUIs which implement Waifu2x out there, I was just mentioning the few I know a little better..

So what is this thread about? Discussing the pros and cons of waifu2x vs other "AI Upscalers" ? Or pros and cons of wafu2x cafe/vulkan/opencl/CPU ? Or is it that a fast avisynth port is not available?

[FranceBB]

So what is this thread about? Discussing the pros and cons of waifu2x vs other "AI Upscalers" ? Or pros and cons of wafu2x cafe/vulkan/opencl/CPU ? Or is it that a fast avisynth port is not available?

I think we're discussing the fact that the only available neural network to uspcale officially supported in Avisynth is NNEDI3 which is quite old (along with an old, non optimized 8bit only version of Waifu2x) and it wouldn't hurt to have a modern version of waifu2x implemented + eventually other neural networks. So we're looking for someone to take up this task and make a new CPU + GPU plugin ported from VapourSynth.

[tormento]

I think we're discussing the fact that the only available neural network to uspcale officially supported in Avisynth is NNEDI3 which is quite old
Sorry for AMD users but CUDA is the easiest path to go. Nvidia did a great job supporting community, not the same can be said of AMD. Perhaps, on a second write, OpenCL could be the evolution but CUDA is so well documented and C++ portable that I can see no other viable option.

[ReinerSchweinlin]

@FranceBB
Thanx, it seems it is the way you say, have been reading the thread once more.. Lots off topic.... Actually, everything is said already, someone simply has to do it ...
@tormento
waif2x already is implemented in many variants and exist as vapoursynth versions. No need for "cuda or not" diskussion... It´s putting it into an avisynth (porting from VS, like France said), what this is about - not the advantages of CUDA:

I have been having good experiences with the vulkan version of waif2x - so I think this version would be the one to go - it´s fast and usable on almost any GPU.

[Reel.Deel]

Anime4KCPP for AviSynth+ (x64 only): https://github.com/TianZerL/Anime4KCPP/releases