stattik_transcode 0.3.1 released

[chud]

i believe there is an error in the script regarding the l5 deinterlacing. the command should be "-vf de/l5" rather than "-vf de-l5". in this case there is no deinterlacing taking place, which is why no filter and "de-l5" result in the same PSNR. given the file naming structure, the "/" in the filter command will mess up the file name. i would suggest just using "-vf l5". the "de" just enforces some mencoder default deblocking. to keep this comparison even leaving it out should be fine.

also, i believe you would see better PSNR results from filmdint with output set to 24fps.

about the PSNR, i got a log file, but i am not sure what i am looking at, and i didn't seem to get an overall PSNR in the mencoder output. must be doing something wrong.

as far as movie dimensions are concerned, there are some general guidelines for the codecs on how many bits per pixel per frame result in acceptable quality (e.g. blocking artifacts). for instance if you want to reduce a video to a particular disk size, you set a bitrate based on the video run time. from this bitrate and fps, you may need to scale down the pixels per frame to land in a reasonable bit/pixel/frame range.

m.

[-stattik-]

I updated the test loop with Chud's correction.

I plan to post more detailed info tomorrow but here's what I got now.

480i

I like pp=md. The PSNR is good and the FPS was about 50% faster than yadif.

1080i

I like pullup. The PSNR was similar with pp=md but pullup was 15% faster.

720p

Everything looked fine and it should have considering the progressive content.


512x384 was chosen to maximize the quality-to-size ratio. Chud explains it well. If you stick with 720x480, you might want to increase the bitrate.

Chud, after each encode, there are 3 lines of psnr info

Code:

xvid:     Min PSNR Y:37.93, Cb:41.79, Cr:45.43, All:39.09 in frame 640
xvid: Average PSNR Y:44.12, Cb:45.17, Cr:47.91, All:[B]44.74[/B] for 720 frames
xvid:     Max PSNR Y:52.60, Cb:50.68, Cr:53.02, All:52.28 in frame 707

I've been looking at the bold number for comparison.

Also, there's dozens of tweaks for these filters for fine tuning, the script takes the basic settings which do fairly well.

[chud]

i think my issues with PSNR output were to do with the mencoder build i was using. first i wasn't getting any summary, and a corrupt file to boot, until i moved the %name% field from being the first argument to the last. as you can imagine that sapped a lot of time tracking that down. then it worked but i got a y u and v summary of SNR, not the nice overall that i saw you had. upgrading to the latest mencoder brought me in line.

well, this has really opened up the ol' can o' worms again on this for me. i went through it all again top to bottom last night. i am transcoding 480i exclusively. the only dint scheme that i didn't have a go at last time was the yadif. i went through a few encodes and personally still like the l5 the best. i find that md and yadif have what look like sharpening artifacts when they trip up, where as l5 does not. the PSNR's are all very similar, and i find that doesn't really reflect seat of the pants evaluation.

for instance, an unscaled encode at the same bitrate results in a dismal PSNR but actually has legs! you are trading off some compression artifacts and encoding speed for some significant picture detail. i would estimate you would need about 2200-2500Kbps for a top quality encode without scaling, but i am going to dig in to do some subjective evlaluations of a single pass unscaled transcode, which takes about the same time as a two pass scaled-down transcode.

i am sure there are as many mencoder settings are there are users, but if you are interested, i would suggest trying a gentler hqdn3d=2:1:0. i am not a big fan of the effects of temporal smoothing so i turn it off completely. btw stattik, what is the fourth value of your hqdn3d? the mencoder man only lists 3...

anyway, a nice little script stattik.

[UncleJohnsBand]

I am going to give the loop another run through tonight.....

Since I am recording via my Hauppage tunners I am not sure if my recordings are interlaced or progressive or something entirely different....they are analog tunners....not digital HD......

[chud]

your PVR150 and 500 are both 480i. same as me...

[UncleJohnsBand]

your PVR150 and 500 are both 480i. same as me...

Thanks....

It appears with the updated sample script where de-l5 has been changed to just l5 that the best PSNR is now between the l5 and the yadif....but I like the l5 results the best.

[chud]

hey stattik, what are you doing with the 1080i content? are you scaling it down to 384 lines? if the original is telecined 24fps content then pullup should be the best choice, but i don't imagine it always is.

anyway, if you are scaling down this 1080i content to <540, you really don't need to deinterlace at all. if you use the "-vf field=0" setting you effectly are only keeping one of the two fields in a frame (interlaced NTSC is 30 FRAMES per second, each frame having two intelaced FIELDS). if you deint and scale down to 540 lines, you are effectively painstakingly combining the two half height fields together into full-height progressive 30fps, and then throwing away every second line to reduce the height by half. using field=0 just keeps only the first field from every frame. what you are left with then is 1920x540 picture or 1440x540 picture at 30fps, so you also have to scale the horizontal resolution to regain the original aspect ratio. this should give you a very good picture at that resolution.

m.

[-stattik-]

To tell you the truth, I don't remember where the fourth value for hqd3d came from. I tried countless different values from around the internet and that one turned out best. My local signal quality can get a little grainy and hqd3d cleaned it up and improved the compressibility.
I plan to add a filter variable to the main script so you can choose the one that works best for you. Perhaps I can add the test loop as well or perhaps it's best to keep that separate.

I scale the 1080i content to 608x336. I don't have a HDTV yet but the transcode beats anything broadcast in 480i. I was thinking of letting Mplayer scan the file and if 24fps progressive is found, to adjust the command line. That probably won't happen for awhile though.

[UncleJohnsBand]

I would keep the two seperate....that way you can change one without worry about impacting something in the other one.

[gEd]

This script looks fantastic. I do have one (potentially stupid) question tho.

Does the edl file contain cut-points and does memcoder use this data to automatically cut the ads when encoding?

I ask as my comskip config doesn't do a good job of finding ads so I don;t want the recording being cut, just transcoded.

If this is the case, what would be the best way of modding the script to not cut the file?

cheers.