Media Developer Solutions – I’m listening

By Craig Hurst (Intel) (7 posts) on June 15, 2009 at 8:21 am

Hello media developers! Let me start by introducing myself to you all. I’m the senior marketing manager and strategist for the media segment within Intel’s Visual Computing Software Division. As I look out over the coming years I’m very excited about Intel’s roadmap for products and technologies. My overall goal is to maximize end user value for our platforms. The single most important element of delivering this value is through the applications that you as media ISVs create and bring to market that showcase that value to end users. So to this end, I am exploring what tools and technologies would most benefit your efforts. I have some ideas, and several things in the pipeline, and will be revealing some of those details over the coming weeks, but frankly I would love to hear your ideas to begin with. What would make your job easier or be most valuable to you as a media developer? Let’s focus for now specifically on challenges with video encode and decode. I’m looking forward to your input!

Categories: Graphics & Media

For more complete information about compiler optimizations, see our Optimization Notice.

Comments (4)

June 15, 2009 4:25 PM PDT

Ajay Mungara (Intel)
Ajay Mungara (Intel)Total Points:
7,924
Community Manager
 Craig,
As a community manager, I am really excited to see that you have the taken the big step of connecting with the community and engaging with the users. I am really looking forward to your future blog posts where you plan to reveal some more details of what is cooking. :)
June 21, 2009 9:52 AM PDT

Igor Levicki
Igor LevickiTotal Points:
16,822
Black Belt
 Well, I have an idea for you.

A lot of video content nowadays is still recorded or produced in interlaced form, which is ill-suited for watching on PC monitor.

Yet, there are no high-quality deinterlacers with real-time performance.

High quality deinterlace requires use of motion estimation. To see how much better image quality is obtained with motion compensated deinterlace check out AviSynth scripts mvbobMod and TempGaussMC.

Unfortunately, the use of ME/MC comes with a high price tag attached -- mvbobMod of a video clip in an NTSC DVD resolution on my Core i7 920 renders at 2.5 FPS without encoding (i.e. saving an uncompressed AVI).

Furthermore, when you want to encode video to x264 after deinterlacing the encoder again has to perform ME/MC step.

So, my idea is -- integrate high-quality, high-speed deinterlacing based on TempGaussMC [http://forum.doom9.org/showthread.php?p=1138514#post1138514] into x264 open source encoder. That way ME/MC won't be performed twice for the same video, and we could all hope to have better deinterlacing quality and encoding performance.
June 29, 2009 4:19 PM PDT

Craig Hurst (Intel)
Craig Hurst (Intel)Total Points:
660
Green Belt
 Igor, thanks for the feedback. This is an interesting idea, and I will have to check out those AviSynth scripts in terms of quality. Makes me think about if there's any merit to using HW accelerated deinterlacing to address performance.
August 28, 2009 2:27 AM PDT

Mikkel Jensen
Mikkel JensenTotal Points:
15
Registered User
 I agree with Igor that deinterlacing in the PC domain lacks proper quality when compared to consumer electronics in general.

And to Craig pushing it onto graphics hardware makes great sense, as it follows decoding the decoded frames are already present in the same domain (memory) as the hardware decoder. it is already being done on major graphics chips (Intel,Nvidia,AMD/ATI), but with limited quality. For transcoding purposes it has so far been hard to exploit as it would require moving it back to CPU for software encoding.

Another issue with PC video playback, which i believe is extremely important in 50Hz countries is some form of automatic switching between refresh rate to lock source and display rate together. If viewing a 50Hz sequence on a 60Hz display every 5th frame has to be repeated, which is very annoying to eye especially on panning scenes, but also motion in general. Though this is mainly a problem of control and setup having impact on OS display architecture and driver models, rather than a issue of computational complexity applicable to a solution thru either optimized software or hardware acceleration, one could go by another approach and use frame rate conversion to interpolate motion compensated frames and go judder free from 50Hz to 60Hz. This could also be exploited to go from 24fps to 60fps. Or up to 120Hz on the LCD monitors that support it. Check out WinDVD9's Digital Natural Motion for the effect. Or any TV marketed as 120Hz/240Hz.

Trackbacks (11)


Leave a comment  

To obtain technical support, please go to Software Support.
Name (required)*

Email (required; will not be displayed on this page)*

Your URL (optional)


Comment*