On Sun, Mar 14, 2004 at 10:40:41PM +0100, Rickard Westman wrote: > Hi, > > I have recently started to transfer some TV recordings onto DVD:s, using > Fredrik Hubinette's mkdvd script. With high bitrates (~ 5.5Mbit/s) the > quality was quite good, but when I tried a lower bitrate (~3.5 Mbit/s), > I ended up with some horrible pulsating DCT blocks in otherwise quiet > (almost still) backgrounds.
Looks exactly like the artifacts I reported last Nov/Dec timeframe that Andrew traced down to an integer overflow problem in the DCT/iDCT routines due to the somewhat limited range of MMX integer operations. I was however pushing it into overflow with very long GOP's, not very high quality values. The solution, at least for the moment until Andrew has time to implement an SSE/SSE2 version of the MMX routines, is to simply "not do that". In your case, don't try to push the -q up that high. > Decreasing the quality setting from the best one (-q 1) to a worse > one (-q 4) produced a dramatic improvement in quality for me. As has been reported way many times on the list, unless you 1) really know what you are doing, and 2) have noiseless sources, you don't want a -q value that high (1-3). It simply causes trouble. Since you say your source is TV, then you do not in any way have noiseless sources, so you don't want to push the -q value that high. You don't say what the rest of your pipeline was, but since the source is TV, you should look into running the stream through yuvdenoise and/or yuvmedian filter as well before feeding it to mpeg2enc. ------------------------------------------------------- This SF.Net email is sponsored by: IBM Linux Tutorials Free Linux tutorial presented by Daniel Robbins, President and CEO of GenToo technologies. Learn everything from fundamentals to system administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click _______________________________________________ Mjpeg-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/mjpeg-users