Requests for the next releases

[Manoman455]

A quick question, in the new version with GPU support will video enhancer utilize all the features implemented on Nvidia's new Fermi architecture?

[Dee Mon]

The plan was to use pixel shaders 3 (which are supported by most video cards now) and to avoid using any super-new APIs which are available on 0,1% of hardware. Even CUDA is a too limiting, not so many cards support it.

[Henery]

My request list:

1) Ability to create filter presets where you can put your favourite filters and/or codec,name it for example "Noisy video enhancing" and then save it.Later it would be easy and fast to edit presets settings and use them in your video with a few mouse clicks.
2)Real time before-after comparison screen like in VDub.It would give some "professional" feel to the software and saves some time.
3)If EEDI2(which is slower than the recent SR in VE) is planned to be implemented in future versions of VEnhancer(according to the other topic),it would be nice to have an option where you can change the up-scaling method between EEDI2 and Lanczos3,so it gives more speed-quality options.
4)Full screen mode.
5)What about audio filter section.Im not sure if there are yet any good filters available but anyway it would be great addition.
6)Softwares logo on the middletop of the window would be nice.Or "Video Enhancer" wrote on stylish font
7)What about creating a Super Resolution processing which covers and prevents compression (deblocking,ghosting etc.) and other artefacts appearing during up-scaling.Software called Cinema HD 2.0 by Engelmann Media has this type of up-scaling called ODU (Object Detection Upscaling).I have used it, and it really makes mobile phone videos look much more cleaner than just deblocking.

Question:Is the GPU acceleration planned to help only Super Resolution processing?What about using it to speed up motion compensation and/or deblocking.

[Dee Mon]

Great list, thanks!
Some parts of it you'll see in nearest release.
EEDI2 is not planned for near future, maybe someday.
Audio... probably in 2.0.
We'll take a look at ODU and see if this approach can be used here.

Yes, GPU usage is planned, this is discussed in neighbor topic here.

[Henery]

I came up with new requests.Of course i dont expect some of them to be implemented in the nearest future releases.

8)Ability to save videos as a "project" files like in Adobe After Effects for example.Then user would later continue to edit the video.
9)I think this could be very useful.Timeline!Then we could add,cut and join multiple audio and video clips.
10)Integrated PSNR meter.Very useful for filter programmers to test their creations quality straight in VE
11)More warning and notification messages.Example,if you accidentally apply Super Resolution on interlaced source without deinterlacing it first,and then start encoding.It would be really helpfull for those who are new to video encoding and dont know anything about deinterlacing,deblocking etc.
12)When starting the program,it would be nice to get automaticly a message which tells the user when there is new version of VE released.

[Dee Mon]

Thanks again!
Here we come to an old dilemma: there are currently two types of users. One type is not so video savvy home users who don't know so much about codecs, formats, filters etc. and just want their video to look better. They need something as simple as possible, and even current form of VE is too complicated for some of them. The other type of users is professionals and geeks who know what is ffdshow and why it sucks (or rocks), what is interlacing and how to deal with it, what is YV12 and why it matters etc. This kind of users want more sophisticated features like audio editing, timeline, avisynth integration and so on.

These two groups of users are very different and currently no one dominates, so VE tries to be good to both (with moderate success). I don't know if it's possible to satisfy both of them. So probably the best way to go is to divide VE into two products: a simple basic version and a Pro version.

Then for Pro version there is danger of becoming one more Premiere or After Effects, very feature rich but too big, too heavy and too complicated for non professionals. Do we really need one more NLE? Maybe just making plugins for already existing big editing apps will be enough...

[Manoman455]

I think separating the program into 2 packages (simple and professional) is a great idea. that way I wont be hindered my limitations set due to the fact that people either have slow computers or don't know how to use this software. Stick them to them to the simple version, we pros will snatch the professional version with its wider range of features and tools. no offense to people who might take offense to what i said, but there are people who wish this program to be more richer in features. That would be me. And the request about automatically checking online for new version of VE proposed by Henry is awesome.

[Henery]

I found an interesting article on wikipedia which is worth of reading and may be better solution to reduce compression artefacts than simple deblocking.Its called Sinc filter (Brickwall filter)
Digital Vision uses this sinc filter in its DVO Brickwall filter.This filter is also in Sonic Cinevision,professional Bluray encoding software,so it must be very powerfull if used correctly.

http://en.wikipedia.org/wiki/Sinc_filter
http://www.digitalvision.se/products/dvo.htm#DVO_Brickwall

[Dee Mon]

Sinc filter is the base for Lancsoz resize which was used in older versions of VE for in-frame upscaling (in v1.9.3 it was replaced by a higher quality one). However it's yet unclear for me how it may be used for reducing compression artefacts. I understand brickwall filter can cut out all high frequencies from the image, this may remove some artifacts but also make picture blurry.

In DVO Brickwall, as I understand, they use the filter before compression to change the set of frequencies, so when a codec compresses the video and quantizes the frequency coefficients, these changes do not produce bad artefacts. Quite clever approach, but it should be used when saving a video, not after it is saved. I think it's possible to make a VD filter which can be inserted at the end of filters sequence to minimize artefacts on saving.

[Manoman455]

I always use Lossless output formats like lagrith and for most others i use H.264. If you really dont want blocks and artifacts i suggest you look into VP7 encoder, or VP8 if you can find it, I use it on anime encoding and it has produced small file sizes and great quality (in my opinion). Hope it helps.

[MC87]

Can I also request CUDA support? Vreveal supports it and the difference it makes is amazing- for one you can actually use your PC while encoding in the background

Also can someone clarify how VideoEnhancers super resolution implementation compares to Vreveals? I know vReveal is severely limited in terms of the types of file superresolution can be used on but say when using SR on a 320x240 video (and turning of other enhancements) which program will give you the better output video? Does one use more reference frames than the other etc.

Thanks

[Manoman455]

I have another request, in future release instead of having buttons that say better quality or better speed you should have it as a slider, one end for high speed and the other for high quality and in between will be the the levels in between the 2. also have more options, as in i would like to select the number of other frames being used to enhance the current frame. Hope i didnt sound rude in this post, i just put what came to mind.

[Henery]

I want to suggest an option to Lanczos3 resize method.Fractals!I think there are no video editing programs which will use fractals,but there is a Photoshop plugin called Genuine Fractals.It uses fractal interpolation to resize the image.The resulting image looks a little bit "fractalised" if zoomed,but still very natural with sharp edges.Fractal algorithms can actually reduce compression artifacts(atleast blocking) just like Lanczos resizer(if used correct parameters).

Then theres another thing in my mind.I cant help thinking if you would create filters of your own.I know that VDub filters are handy and some of them are rather good,but imagine that you could utilize multi-frame analysis and motion compensation for example on sharpening or grain management filter for better results.Your SR upscaling is awesome and i think your own filters would be too.I hope you get more manpower somewhere to speed up future releases developing.

[Henery]

I have a new suggestion to the Super-Resolution.What if we could use vectors on upscaling.All of you propably know that vector graphics can be enlarged infinitely without losing any detail and also edges stay sharp.In my opinion,vectors could be a solution to the two main problems of the recent version of VE Super-Resolution.

-Strict 2x enlargement factor.Unless there is still something why vectors cant help here.Otherwise we would get completely rid of the enlargement limit and upscale straight from 480p to 1080p without quality loss.
-Aliasing free edges.That way we dont have to use separate antialiasing filter.

There are two Photoshop filters of this type:Alienskin BlowUp and Imageners Kneson unlimited++ interpolation.Ive only tested the first.Blowup does excellent job even on 4x enlargements .Blowup also uses vectors to images original textures,which makes them sometimes look unnatural.If VE uses this technology,i would suggest using it only for edges.That way we would maintain images/frames natural textures.The other details inside edges would be upscaled using fractals,because they can maintain small size details.Noise and compression artifacts immunisation combined with those technologies would provide great quality.

Here is Blowup vs Bicubic comparison.Pay attention especially to the cartoon and Eiffel tower examples edges.
http://www.alienskin.com/blowup/blowup_examples.aspx

Imagener Kneson Unlimited++
http://www.imagener.com/unlimited.html

If there are any reasons why VE cant utilize vectors,then let me know.Oh,and by the way,how are the plans of GPU implementation going?

[Dee Mon]

Thanks!

Fractals are very time consuming to work well. For each part of the image we need to find a similar larger block, this is done by trying hundreds of variants each time doing downscale and pixelwise compare, this is really CPU intensive. And results are not always good because this is just cheating: from math point of view details you get with this method have nothing to do with real picture details, they just may look nice. Just because photos and video frames do not really have fractal nature.

As for vectors, it's true that if you've got a vector image (like a TrueType font, for example) you can scale it as you want without losing quality. The problem is that input video is not vector, it is raster, bitmap. And the process of converting bitmap to vector is not easy at all.

GPU is still somewhere in future, sorry.