Saturday, February 28, 2004

Smart Smoother HiQ Speed

Here are some more notes about processing speed with the Smart Smoother High Quality filter in VirtualDub (previous post). These are for 720x480 30fps captures on an AthlonXP 2600+ system using PC2700 DDR memory. Amount is always 254 and all were done as "weighted with difference", other values are r=radius, t=threshold, m=maintain. The last number is the amount of minutes required to encode a single minute of source material.

r9 t30 m15 12.0x
r9 t15 m10 11.0x
r9 t10 m5 9.5x

r7 t60 m30 7.5x
r7 t30 m15 7.0x
r7 t15 m10 6.5x
r7 t10 m5 6.0x

r5 t40 m20 3.5x
r5 t30 m15 3.5x
r5 t20 m10 3.5x

Notes:

- r7 t60 m30 looked blurry, losing fine detail in areas such as a shot of the snowfield

posted by Wuphon's at 1:01 PM

Thursday, February 26, 2004

A7N8X and GeiL DDR Memory

I've been getting a lot of random reboots lately, so it looks like the Asus BIOS 1007 didn't completely solve my problem. (It's better then it was.)

So... for my next trick, I went into the memory settings in the BIOS and changed from CL 2 to CL 2.5. I still expect that I'm going to have to buy new memory (probably Kingston), but since cash is tight at the moment I'm trying other avenues first. (Yes, I know some would say that this should've been an obvious idea that I should've tried 10 months ago...)

The downside of all this is that I can't use the AthlonXP 2600+ box to do any intensive VDub or TMPGEnc work. TMPGEnc seems to be very sensitive and throws errors often, VDub seems to be a bit more forgiving. Of course, when you bomb out near the end of a 4hour conversion, it's rather frustrating. It also means that since I only have my P4 1.4Ghz laptop CPU to work with, that I'm currently running behind on my conversion queue (I have a 2hour clip that is taking 20 hours to filter in VDub). The Athlon box is around 30-50% faster then the P4 laptop, so if I can get it stable I'll do better then double my processing capacity.

(Update +20h: System has been up and running all day, encoding MPEG2 and doing VDub filtering with zero errors and zero reboots... keeping my fingers crossed.)

posted by Wuphon's at 11:46 PM

Smart Smoother HQ Rendering Speed

Here are some rendering speeds for a 352x480 30fps clip in VirtuaDub when using various filters (examples of effects). All processing was done on an AthlonXP 2600+ system with DDR PC2700 memory.

Temporal Smoother - constant 48fps, regardless of what setting was used (tested at 2/4/6/8 values).

Smarth Smoother High Quality - Radius setting has the most effect, average vs weighted has little effect (roughly 0.5fps difference), difference vs no-difference has little effect. Threshold has the 2nd largest impact on processing speed (radius setting has the largest impact).

r05/t005 - 22fps
r05/t010 - 20fps
r05/t020 - 19fps
r07/t005 - 14fps
r07/t010 - 11fps
r07/t020 - 9fps
r13/t005 - 6fps
r13/t010 - 4fps
r13/t020 - 3fps

For live-action, I like "r5 t10 a254 m5 weighted w/ difference". For cartoons with large areas of flat-color, "r7-9 t10-15 a254 m5-10 average w/ difference" is my preferred (depends on the source).

posted by Wuphon's at 11:42 PM

TV Episode Capture to DVD

Switching up my methodology a bit. Previously, I was capturing at 720x480 30fps or 352x480 30fps, encoding at 6000kbps (video) and 256kbps (AC3 audio) which allows me to fit 2 hour-long TV episodes per DVD if the commercials are removed. (Roughly 48min per episode, 96 min per disc. DVD capacity by bitrate table)

However, 6000kbps for half-D1 is over-kill so I've been doing some testing to see how low I can go before I start noticiing artifacts. First off, I've switched completely to half-D1 (352x480 30fps) instead of capturing at full-D1 (720x480 30fps). I tried encodes at 3000, 4000 and 6000kbps and really couldn't see much artifacting even at 3000kbps from a slightly-noisy VHS source. So while I'm not entirely sure that I can go as low as 3000, it's possible that I can go that low or even a bit lower.

My first test is (3) episodes encoded at 3800kbps. This created 3.66Gb of VOBs when I authored the disc, with a best-case of 3.48Gb and a worst-case of 3.78Gb. Assuming disc capacity of 4.30Gb (allowing extra space for PAR2 files), my test used up between 81% and 88% of the available disc space. Even though I told it to encode at 3800kbps for the video, actual average bitrate works out to only be 3500kbps (due to use VBR). With a bit of math (and some hand-waving) that means I could probably boost the video bitrate to as high as 4000kbps (plus 256kbps AC3 audio) and still fit (3) episodes per disc. (Update: the 3800kbps test disc looked fine given the quality of the source material.)

My next test will be to encode (4) episodes for a single disc (192 minutes). This gives me a bitrate budget of 3170kbps, minus 256kbps for audio, leaving me with 2900kbps for video. That seems attainable since my test clips still looked okay at 3000kbps. Tempting, since it will also cut my DVD-R usage even more.

I'm also filtering the video using VirtualDub to remove a bit of noise. The filter is "Smart Smoother High Quality" with settings of "radius 5 threshold 15 amount 254 maintain 5 weighted with difference". I may back that off to threshold 10 to lower the smothing effect just a hair.

posted by Wuphon's at 7:56 PM

MJPEG Capture Data-Rates and Quality

Did some test capture off some VHS source material to check data rates (earlier listing and recent list). Capture settings were 720x480 30fps with 48kHz 16bit stereo audio.

RAW - 20Mb/sec - 72 Gb/hr
Q20 - 7.8-8.5 Mb/sec - 27.4-29.9 Gb/hr
Q19 - 2.6-3.2 Mb/sec - 9.1-11.2 Gb/hr
Q18 - 1.7-2.1 Mb/sec - 6.0- 7.4 Gb/hr
Q17 - 1.5-1.7 Mb/sec - 5.3-6.0 Gb/hr
Q16 - 1.2-1.4 Mb/sec - 4.2-4.9 Gb/hr
Q10 - 0.7-0.9 Mb/sec - 2.5-3.2 Gb/hr

At Q10, it was very easy to see pixelation around things like text or hard edges on the screen. Q16 and Q17 were pretty good, but with a semi-noisy VHS source it's hard to see a difference between that setting and a higher quality setting. Normally, I capture at Q19, which is a good trade-off between space/quality. To save more space and to make encoding easier, I can capture at 352x480 which chops my data rate by roughly 30-40%.

If I was doing a PVR-type setup where I knew that I'd be processing the video prior to putting it out to DVD (which means I don't want to capture straight to MPEG2), I'd consider Q17 or Q18 as a way to save even more space.

Right now, I'm using (5) VHS tapes to do my weekly TV episode captures (8 hrs/wk or so)... which is not quite enough for me if I should screw up a capture and not catch it until after I make the DVD at the end of the process. Usually, I do the capture / encode / burn around 3-5 days after I record the episode off the air. That process can take 24-48 hours depending on what I have in the filtering / encoding queue. A more comfortable buffer would be (15) tapes where I'd only use each tape around once every 3 weeks, giving me almost 2 weeks of buffer time to go back and catch a goof.

Why figure that out? Because I'm trying to estimate disk requirements should I switch to recording straight to disk. 30 hours at Q19 is 315Gb (ouch) and that needs to be multiplied by 2 with an additional 4Gb/hr of scratch space to account for working files. (30 * 10.5Gb/hr * 2 + 4 * 30 = 750Gb) Switching down to Q18 and 352x480 (4.5Gb/hr) cuts that about in half (30 * 4.5 * 2 + 4 * 30 = 390Gb). Still a heck of a lot of space, but finished clips could be off-loaded to a 2nd disk.

posted by Wuphon's at 12:28 AM