Requests for the next releases

[Henery]

Since Super-Resolution is using multiframe analysis,would it be possible to implement spatio-temporal frame analysing methods in it?I have read and seen many spatio-temporal noise reduction filters and they are told to produce great results of removing noise while keeping details.VReveals SR smooths some of the artifacts away but you have sometimes sacrifice lot of details.If VE Super-Resolution uses this spatio-temporal methods with succes,it would upscale the real details and edges while avoid enlarging the noise and maybe other artifacts too

If VEs motion-compensation algorithm stabilises the motion of the whole scene,then i have an improvement idea.What if the motion-compensation engine would detect that what kind of movement there are in the frame.So instead of using one compensation method for every motion,it would use several different algorithms suited for different kinds of motion.If the background is static,we dont process it at all and that should save some time.There are also better methods to estimate the motion than those block based or motion vectors.I cant remember them now but i have seen PDFs of them somewhere.

And last but not least,does the next version of Super-Resolution have any improvement on edge upscaling?Recent version of SR is otherwise good,but sometimes edges are badly aliased.

[Dee Mon]

Our current SR method considers input as proper signal (not noise) and this is the assumption all algorithm is based on. Particularly, when upsizing 2x each pixel becomes a quad of 2x2 pixels and their average value is equal to the value of the input pixel. And if you just average each 2x2 block back to 1 pixel you'll have your original video. This means that in current approach all noise is preserved. Same principle works with aliasing: if input video is aliased, output video most probably will be too.

There is indeed a way to blend several motion compensated frames to remove noise, but usually it leads to smooth picture with less details. Removing noise and keeping (or even increasing) details at the same time is very hard.

Our motion compensation method is already block based, when for each block a motion vector is found pointing to a block of same size in previous frame that is most similar to the given block. Searching for such blocks is the most computationally expensive part of the algorithm, this is where most of the time is spent.

[Sunny666]

Ди Мон, когда уже выйдет новая версия? чтото вы долго марафет наводите)

[Dee Mon]

Скоро.

[Robien]

preview option does not work with vd filters i have difficulty to apply the filter could you support it please

[Dee Mon]

That's planned for version 2.0.

[Henery]

I just saw the before/after comparison animation of VE on its homepage.Would it be possible on future versions of Video Enhancer have an preview window with draggable line showing both before and after results.That would be so handy and time saving addition.
Another suggestion:I have long list of filters installed and sometimes it is frustrating to scroll down the list just to choose,for example two of MSUs filters separately.The scroll would atleast stay in that position where it was after clicking OK or,the filter list would be maximized to whole screen height to show more filters at once.
Is there any way for us users to help on the user interface developing,than posting comments?Maybe you could sometime post here few screenshots of new UI´s.

[Dee Mon]

Our plans for version 2.0 include one big video window where filters (and SR) are applied like different layers in Photoshop: you can enable/disable any of them and instantly see the changes. That would be more handy than dragging some line.

As for the filters in version 1.9.6 we're going to split them into categories to make search and selection much easier. Also Video Enhancer will contain database of all filters and will be able to download and install required filters itself.

I guess we could post some in-progress screenshots here so people can participate in the process.

[Manoman455]

Wow I just thought of something that I should have addressed so long ago. With the increase in more people transitioning to x64 operating systems, will there be a 64 bit version in the near releases? this would allow more memory to be accessed and used if I'm correct and surely benefit some aspect or video enhancer. Just a great thought I had, or I think it was great.

[Dee Mon]

Actually going 64 bit will add nothing but troubles: Video Enhancer uses MS DirectShow to combine all required decoders, filters and encoders in one chain (a graph really), and all major codecs are 32-bit and can't be used in a 64-bit application. And all VD filters are 32-bit, using them from a 64-bit process is a big problem.

As for the memory, 64 bits give advantage when there is a need to use more than 2 GB of RAM in one application, but VE doesn't need so much memory: when it works it usually takes less than 100 MB. Thanks to streaming nature of all the algorithms involved, at each moment only a few uncompressed video frames must be kept in memory.

[themaster1]

As for the filters in version 1.9.6 we're going to split them into categories to make search and selection much easier. Also Video Enhancer will contain database of all filters and will be able to download and install required filters itself.

I have a suggestion for the vdub filters: it would be very useful to know which ones work in YV12, i know all of them should work in RGB but some work in YV12 i believe but you never really know until you test yourself and it works..or not (time consuming) like i've discovered with the MSU filter recently...

Just a thought

On an unrelated note, do you know a filter that can achieve the same results, see here (vdub or avisynth) that would be a nice plus to use in conjonction with VE.

[Dee Mon]

According to our data from last year almost none of external VD filters support YV12.

No, I don't remember any available filter like super-precision. However recently  I had some thoughts on making a similar filter for brightness correction. When a very dark video is made brighter, its histogram looks like the ones shown on super-precision filter page, so a need for such an algorithm naturally arises.

[Dee Mon]

Version 1.9.6 is out.
Main changes:
1. Super Resolution speed increased 1.5 - 2.2 times!
Now on a Quad Core 2.33 GHz we get following speed:
640x360 -> 1280x720: 15-25 frames per second in RGB mode (depending on quality settings) and 23-37 FPS in YV12 mode.
960x540 -> 1920x1080: 6-11 FPS in RGB, 11-20 FPS in YV12.
To make it possible we used SSE2 heavily, so you need a Pentium 4 or later CPU. Practically almost any CPU made after year 2000 is ok. Previous century machines are not supported anymore.

2. Higher SR quality, especially in RGB. Less artifacts on borders and edges of moving objects.

3. One more option: Show SR mode. Now it's easy to see in what mode (RGB/YV12, size, quality) each of SR inctances work.

4. Seven different languages of user interface. Added German, Spanish, Portuguese (Euro and Brazil variants separately) and Greek. French is coming soon.

[MC87]

Does anyone else have their video playback get accelerated at the beginning of videos? Sound also seems to go out of sync quite a lot.

I use Xvid for video encoding but do not re-encode the audio stream (leave it as is) could that be causing the audio sync problem?

Anyone know what a good setting for Xvid is? I'm using Target quantizer 2 (should I use 3 instead?) - it's a pity Xvid doesnt do VBR single pass compression like WMV can do (like in vReveal).

Would it be possible to include a ready to use WMV encoder for the next version of VE?

[Dee Mon]

Currently VE writes AVI files, and AVI always has a constant framerate. Unfortunately there is no way to reliably determine framerate of source video, so VE has to guess by first 20-30 frames. In most cases it works fine, but with some source files it may cause problems. And incorrect framerate leads to audio/video out of sync.
There are no such problems in WMV since it's not limited by constant framerate.
Yes, we're going to add WMV output in nearest updates.

When working with xvid and divx I usually set some bitrate, not quantizer. Good bitrate depends on video dimensions and framerate, also on what I'm going to do with this video (e.g. play at home or send to youtube), so I can't tell a universally good setting.