tag:blogger.com,1999:blog-745641191327454996.post4753542765723988213..comments2022-07-19T02:24:22.750-07:00Comments on Max Sharabayko Blog: Next Generation Video Codecs: HEVC, VP9, DaalaUnknownnoreply@blogger.comBlogger16125tag:blogger.com,1999:blog-745641191327454996.post-1841276335997963442014-05-16T22:05:03.182-07:002014-05-16T22:05:03.182-07:00Raul,
As far as I know there is no human psycho-vi...Raul,<br />As far as I know there is no human psycho-visual perception model so far. Therefore neither SSIM, nor PSNR can be considered good distortion metric: there is just no criterion of truth.<br />However I do agree that to be confident in RD-performance results, VQM or SSIM should be used also. I guess I'll update the comparison later on, but don't believe the results would change.Max Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-91477872177790105012014-05-16T08:51:45.342-07:002014-05-16T08:51:45.342-07:00Your comparison is based on PSNR values, something...Your comparison is based on PSNR values, something which Daala doesn't target as a performance factor.<br />Actual perception is what Daala is targeting at.Podersupervacahttps://www.blogger.com/profile/03203643627188416109noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-29306585034671031502014-05-15T20:36:37.207-07:002014-05-15T20:36:37.207-07:00In the "See Also" section I sahre a link...In the "See Also" section I sahre a link on such comparisson to x264. Though I think it's not a valid comparisson prejudiced.<br />Anyway do you suggest that x264 in constant quantizer or constant quality mode would outperform JM?Max Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-76925041103661251352014-05-15T08:32:38.275-07:002014-05-15T08:32:38.275-07:00comparison to x264 please? :)comparison to x264 please? :)Roger Packhttps://www.blogger.com/profile/01578246846716577925noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-83199669050903257182014-04-13T20:20:45.242-07:002014-04-13T20:20:45.242-07:00Well, Shruti, there are at least two ays to compar...Well, Shruti, there are at least two ays to compare pure encoder performance without rate control etc. One is to compare in constant quantization mode, another is - in constant quality mode. I prefer the first one as more simple, involving less decision-making algorithms.<br />So yeah, I suggest max-q=min-q=QMax Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-60218170425940827702014-04-05T14:13:21.272-07:002014-04-05T14:13:21.272-07:00Incase I am comparing a particular sequence for in...Incase I am comparing a particular sequence for intra-frame comparison of HEVC and VP9, The Quantisation parameter will have to be set to a value same as in HEVC. In such a case do you suggest me to use max-q min-q =Q?? what would be the best usage in case of Intra frame comaprison.. Could you please help me with this?Anonymoushttps://www.blogger.com/profile/11497143358047146042noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-56949484082351946492014-03-26T12:17:47.583-07:002014-03-26T12:17:47.583-07:00That still asks for a username and a password. That still asks for a username and a password. spidyhttps://www.blogger.com/profile/06366649256222658467noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-30491196156117917872014-03-26T08:27:20.910-07:002014-03-26T08:27:20.910-07:00Hello.
Try the following link: ftp://hvc:US88Hula@...Hello.<br />Try the following link: ftp://hvc:US88Hula@ftp.tnt.uni-hannover.de/testsequencesMax Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-8374936003212781122014-03-26T08:17:29.589-07:002014-03-26T08:17:29.589-07:00Hi I'm trying to download the test sequences a...Hi I'm trying to download the test sequences and they keep asking for a username and a password. Any idea where to get that?spidyhttps://www.blogger.com/profile/06366649256222658467noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-5309171723152194562013-11-14T22:07:21.387-08:002013-11-14T22:07:21.387-08:00http://iphome.hhi.de/wiegand/assets/pdfs/2012_12_I...http://iphome.hhi.de/wiegand/assets/pdfs/2012_12_IEEE-HEVC-Overview.pdfMax Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-30087292342930434412013-11-14T21:38:24.146-08:002013-11-14T21:38:24.146-08:00Hi Debargah, I am unable to follow HEVC standard e...Hi Debargah, I am unable to follow HEVC standard even after reading documents. Can you provide me any source with good Explanation. ThanksAnonymoushttps://www.blogger.com/profile/15001605121642202155noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-41705036224165139342013-11-13T00:02:07.727-08:002013-11-13T00:02:07.727-08:00The "constant quality" mode (end-usage=3...The "constant quality" mode (end-usage=3) in VP9 ensures that all regular INTER frames use the quantizer Q specified in the cq-level=Q parameter. The key frames and alt-ref frames however will use lower quantizer (higher quality). <br />On the other hand, making max-q = min-q = Q, will make all frames have the same Q, which is not the best in coding performance. Just wanted to clarify in case there is any confusion.Anonymoushttps://www.blogger.com/profile/17678072972364888219noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-14603009780562937402013-11-04T20:09:19.029-08:002013-11-04T20:09:19.029-08:00This is true for constant quality, but the tests w...This is true for constant quality, but the tests were held in constant quantization mode. I consider this more fair comparison.Max Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-57798791395386852792013-11-04T09:30:58.477-08:002013-11-04T09:30:58.477-08:00For 'constant quality' encoding, a better ...For 'constant quality' encoding, a better option is to run with:<br />--end-usage=3 --cq-level=%Q% --kf-min-dist=%KF_DIST% --kf-max-dist=%KF_DIST% ...<br />Also, currently the 1-pass encoding option in libvpx is kind of broken. So it is better to test with 2 passes for now - even though I agree that makes a fair comparison with the HM difficult.Anonymoushttps://www.blogger.com/profile/17678072972364888219noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-89140825538646813142013-10-29T02:59:14.382-07:002013-10-29T02:59:14.382-07:00VP9 encoder version v1.2.0-4614-gc5e9108 was run w...VP9 encoder version v1.2.0-4614-gc5e9108 was run with:<br />--min-q=%Q% --max-q=%Q% --kf-min-dist=%KF_DIST% --kf-max-dist=%KF_DIST% --passes=1<br />where %Q% - is a quantization value, %KF_DIST% - key frame distance valueMax Sharabaykohttps://www.blogger.com/profile/15800163406063795337noreply@blogger.comtag:blogger.com,1999:blog-745641191327454996.post-5781575277643367782013-10-28T09:36:22.322-07:002013-10-28T09:36:22.322-07:00Thanks for conducting the tests. Wold you mind sha...Thanks for conducting the tests. Wold you mind sharing the following information about how VP9 was run?<br />In particular, what was the hash or at least the rough date of the code that was used? and more importantly, what the command-line was ?Anonymoushttps://www.blogger.com/profile/17678072972364888219noreply@blogger.com