[bry4ntt]

I wanted to reconvert/resize all my HQ 700mb .avi files to MQ 350mb. I managed to do a two-pass encoding using virtualdubmod and xvid codec (full quality 1st pass)but the quality seems so much different, the pixels are sometimes really noticeable and blurry. I know that re encoding reduces quality, but in my case the quality is is much much lower than what I expected (I had some MQ 350mb files downloaded and the quality are much better than my new reconverted files).

My question, is there any way to reconvert without so much quality loss? I heard about AutoGK, will it be better if I use it? I don't really trust the automatic-software thingy..

[DaRkViEt73]

Maybe you're asking for too much. Afterall, those MQ 350mb files you're downloading were probably encoded DIRECTLY into that size thus minimising alot of quality loss. Since I don't have these files I can't be sure... but maybe they're also encoded with less resolution (like 512x384 instead of 640x480 or whatever). That would allow higher bitrates for a given filesize thus increasing quality at the expense of resolution. Remember you're cutting the filesize in half... you'll be lucky that the video quality doesn't go down by half either.

In any case, I do have some suggestions. The first thing is we need to make sure you are maximising the quality of your encodes. http://www.soompi.com/forums/index.php?s=&...t&p=2462432 Look at that post and set your first and second pass according to what I've specified for each pass. In your situation I'm guessing you'll probably want to set your Quantizer=4.00 instead of 3.00 like what I've got however. On the second pass, use Weight=1.00 instead of Quantizer. Do the other settings like I've posted but instead of using 50-70% of the number shown, input your target filesize as 350000 for 350mb.

Before you actually do the first and second pass, however, you should use the temporal smoother filter and set the field strength to level 3 (4th notch from the left). This will smooth out the picture, reduce video noise, gets rid of alot of macro-blocking and overall it will increase quality in your situation. Simply add the filter before you start your first & second pass and it will do it's magic. You may also want to consider reducing the resolution a bit if the final quality is still not satisfactory for you. There really isn't much else I can recommend to maximize quality. Well actually there IS but let's not go there.

[bry4ntt]

hi DaRkViEt73,

Thank you for all your explanations and recommendations..

I tried to re-encode a file again using your methods.. but there isn't much improvement in quality.. the quality is still grainy and blurry.. will it be better if I run the job in fast recompress mode? I remember reading in a forum sometime ago that a compressed file shouldn't go on full processing mode, is it true?

[DaRkViEt73]

You can try it and see if it helps. In theory it's better to use Fast Recompress if your source is an xvid file but you give up the ability to use virtualdub filters which means you must use AVISynth to apply filters. AVISynth is more complicated and intended for advanced users so you'll need to read up on it.

Again, don't expect incredible results since you're trying to get similar quality in HALF the space which just isn't gonna happen. In your case, you will have to compromise and I can offer some further suggestions.

1) The most obvious is to do your encodes directly from a better source than that 700mb file. But I think it's safe to assume you can't do this or we wouldn't have this thread.

2) Use a different codec. H264 is more efficient than XviD and retains more details but requires over twice the encoding time. Or use Windows Movie Maker and encode into WMV or ASF formats [gawd I hate these two formats tho].

3) Resize the video down to a lower resolution. If the original XviD file is in 640x480, then resize your video to something like 360x480 or 480x480. Then set XviD's aspect ratio to 4:3.



Doing this will save you significant space. It's harder for people to recognise a loss of horizontal resolution versus vertical resolution so this is a great way to increase your quality. However, some movie players may ignore the 4:3 AR flag set in the file so your video may not look right. VLC Player and Media Player Classic both recognise the 4:3AR flag so the video will be sized correctly when played back.

4) If you want to stick with XviD, go download EQM's v3 Ultra Low Rate custom matrices and use that instead of H263 for your quantization type.



Just in case you can't find it, simply go there and enter in all the values you see in the picture above. That is EQM's v3ulr custom matrix which is geared towards encodes like what you're trying to achieve at 350mb. You will gain more sharpness than using the H263 matrix.

5) Use the brightness/contrast filter in Virtualdub. Set it on -6% brightness and +6% contrast. Video codecs like XviD love darkness and compress much better when things are a bit darker. You can potentially see a significant quality increase from this. The compromise being everything's a bit darker but you get better contrast so you may like the results.

6) In XviD, change your Maximum I-frame interval to 600. This will affect your ability to seek throughout the video. What you will notice is when you skip ahead, it'll jump ahead more than normal [potentially up to 20 seconds at a time]. Keyframes are now created up to a maximum of 20 seconds apart instead of the usual 10. It could, however, gain you a little bit more quality due to less keyframes being created. I doubt you will see much improvement.

7) Convert the video into 24 frames per second instead of the normal 30. You will gain about 20% more quality but depending on the video, you may see alot of stuttering. However, it's generally not a good idea to do this on video someone has already encoded like downloaded XviD files and stuff.


Again, keep in mind you're asking to do the impossible so don't expect the quality to be much better than what you're getting now. Option 1 is really the best way to get decent encodes at that filesize and that's how those MQ 350mb files were created in the first place. My last piece of advice is you may want to consider not using the temporal smoother if you're gonna resize to a lower resolution or do one of the other tricks that save significant space. Generally, the less filters you need to use the better. Have fun!

[awdark]

._. I am always impressed by your replies to these video posts, and that is why I usually just leave them to you.

Personally, I just use pocket divx encoder to shrink things lol

This thread is insanely advanced and I just know how to use the wizards ill read this stuff later and hope to make nicer vcds or something in the future

[happyroach]

Hahahahaha I'm sooo bookmarking this.

Not that I worry much about space, but retaining more quality without increasing file size is pretty awesome.

Now I just need to give H264 another try...it takes bloody long on core 2 duo as well >:[

And isn't it kind of useless to run duo pass on a pre-encoded xvid?

[DaRkViEt73]

And isn't it kind of useless to run duo pass on a pre-encoded xvid?

No because you're still trying to hit a certain filesize. Regardless if it's a prerecorded xvid file, a DVD rip or TV/Satellite capture, DivX/XviD still needs to know the characteristics of the video first... hence an information gathering 1st pass and actual encoding 2nd pass are still required for best results and accurate filesizes.

Now I just need to give H264 another try...it takes bloody long on core 2 duo as well >:[

Overclocking your C2D to over 3.2ghz-3.6ghz helps.

[happyroach]

Ah I see. Well I hardly re-encode xvid/divx files so I always duo-pass my transport stream stuff heh <3

But EH HOW DO I OVERCLOCK THIS BABY. I overclocked my old desktop (athlon xp) fine but this one it looks like the BIOS is locked in some places and I haven't been able to overclock anything. (I have a Dell Inspiron with Core 2 Duo T5600)

Hahahaha I HAVEN'T FORUM-ED FOR TOO LONG >:D

[DaRkViEt73]

But EH HOW DO I OVERCLOCK THIS BABY. I overclocked my old desktop (athlon xp) fine but this one it looks like the BIOS is locked in some places and I haven't been able to overclock anything. (I have a Dell Inspiron with Core 2 Duo T5600)

That is for another thread and hopefully a few people will jump in to help you with it [assuming it's possible to do at all on a Dell]. Boka might get the itch to hit that warn button otherwise...

[bry4ntt]

thank you so muchhh for the tips.. i'll try again tonite.. i usually do the encoding when i'm sleeping.. my comp is so slow and I can't afford to wait a few hours...

I guess I'll have to compromise with the quality since I don't have the source file.. I did try using EQM's v3 Low Rate (not the ultra low rate).. I'll try to use the ULR one tonite.. btw.. do you use the matrices on both 1st pass and 2nd pass? (I used in on both)

I'll also try using Fast Recompress.. see if it works..

[DaRkViEt73]

btw.. do you use the matrices on both 1st pass and 2nd pass? (I used in on both)

yes you should use the same matrix on both 1st and 2nd pass for it to be accurate. You may want to experiment with different matrices as well to see what result you like best. Once you nail down the ones that give the best result in this scenario the next time you do this type of encode it'll be much easier.

Remember that you may want to adjust your 1st pass's quantizer if quantizer 4 is giving you too small or too big a filesize. You want the 1st pass to be at least equal to if not somewhat bigger than 350mb. In your scenario, taking into consideration the settings you'll be using, you want the 1st pass to be between 350mb min and 700mb max [since 50% of 700mb = 350mb]. The closer the 1st pass is to 350mb without going UNDER, the more optimized your encoding is. You generally don't want it to be more than about 500mb tho. Using the EQMv3ULR or EQMv3LR matrices may require you to decrease the quant from 4 to preferably 3.00 instead or even as low as 2.00 for maximum quality if the 1st pass filesize is too low. Generally the higher the quantizer number, the crappier it looks [especially with such low bitrate quant matrices] but the lower the filesizes are. You have to find the balance and that only comes with experimentation and experience.

Remember that as a general rule, the higher bitrate matrices are more flexible when it comes to the "50-70% of the 1st pass filesize" rule... meaning you can do as low as 50% of the 1st pass filesize for your 2nd pass and it'll still turn out looking good. With the lower bitrate matrices, it generally requires 70-80%+ of the 1st pass filesize to get satisfactory results.