Question to Encoders: File Sizes getting out of hand?

[comatose]

hehe <message too short>

[Waryas]

Originally Posted by comatose

I noticed that grainy anime tends to still look good at CRF 22. Yeah, there are differences between that and 19, but I'm not encoding a Blu-ray =/

There are differences between 16 & 18, yeah... but it's the kind of difference that doesn't really matter. 18 is almost always beautiful. You kinda hit diminishing returns then... you need to pump in tons of bits for small changes.

Check natsu no arashi 01 .ts if you want an encoder-hell source.
It hit 460MB on crf 18 @ 848x480 without any filtering (Except IVTC, it's VFR)

[Kristen]

OK, made 2 encodes for Zettai Karen Children 05. One was at 350 MB which generated an average p-quant of 18.6. The second was at 700 MB which generated an average p-quant of 14.6. Both were encoded from the same raw, same filters, and same x264 avisynth script. The only difference is the bitrate.

700 MB:
350 MB:

[Tofusensei]

Identical, for all practical purposes.

Can you try a 175 meg encode? I'd curious how that stacks up. 350 vs 700 is a little unrealistic.

[Kristen]

Originally Posted by Tofusensei

Identical, for all practical purposes.

Can you try a 175 meg encode? I'd curious how that stacks up. 350 vs 700 is a little unrealistic.

I can try, though it'll take 4 hours to make, and I sleep in 3, so I won't have those comparisons until tomorrow. I probably should make a 233 while I'm at it (Old size HD) to compare as well.


By the way, this show is high action, high motion, 29.970 fps, HD, and 24 minutes 30 seconds long, so it should be perfect for these types of comparisons in my opinion, since it is so poorly compressible. >.< If there's no real practical difference here, I don't see why ANY regular 24 minute 23.976 fps show would warrant more than 350 MB, unless it was a Blu-Ray.

[Tofusensei]

Thanks, I'd appreciate.

What I meant was, we all knew there wasn't going to be any real improvement from 350 to 700, which is why 175 (and yes, 233) would be good data. Thanks ^^;

[Kristen]

Originally Posted by Tofusensei

Thanks, I'd appreciate.

What I meant was, we all knew there wasn't going to be any real improvement from 350 to 700, which is why 175 (and yes, 233) would be good data. Thanks ^^;

Hehe, yeah. I do suspect that there will be some blocking at 233 MB, since Chihiro got yelled at a lot for doing 233 MB for some episodes when it needed more. At 175, most likely.

But the screenshots I posted were on the thread topic of file sizes which go over 350 MB.

Well, only the encode will tell us for sure. XD At least I remembered to put the frame numbers in the pictures.

[martino]

Originally Posted by Kristen

By the way, this show is high action, high motion, 29.970 fps, HD, and 24 minutes 30 seconds long, so it should be perfect for these types of comparisons in my opinion, since it is so poorly compressible.

Doesn't seem to have much detail though... and how about some dark scenes?

[Waryas]

Originally Posted by Kristen

I'm sorry, not to troll or anything but can i ask what kind of filtering you used ?
The banding on the last panel looks pretty bad.
Can you post the source, UNFILTERED ? (just ivtc + resize/crop, nothing else)

so yea, if you're going to oversmooth and use LSF (thus introduce banding), you're gonna get the same shitty result whatever the size because the video already lost details (Gradient in this case).

On the other hand, if you kept the filtering minimal and actually added a debander to your script then the 700MB would look MUCH better.

There's an appropriate time for each size and it depends on your filtering : If you oversmooth, low size will usually do the trick and this is exactly what happened in this case.

[DryFire]

Sadly, you can't see temporal noise in screen shot. Something that goes from 24 or so to 18 would be more interesting.

[mieruno]

@ op 1gb of hdd space is less than 3cents, if you can't afford that i recommend you go stream your anime

i mean everyone else is willing to download stuff regardless of the filesize!

[Zalis]

Originally Posted by mentar

Sure I do. Season 1 was based on Raws, in other words, on sources which were already recompressed, and back in those days, mostly recompressed with ASP codecs (XviD).

Ah, so it's an issue of different/improved sources between the two seasons. Thanks for the explanation, I'm sure I'll find a way to survive the increase.

[Arm]

Originally Posted by Kristen

If there's no real practical difference here, I don't see why ANY regular 24 minute 23.976 fps show would warrant more than 350 MB, unless it was a Blu-Ray.

Take a look at Souten Kouro. I'm no encoder, but he said it's a good example of a 340mb release not being enough.

[Dark Shikari]

Originally Posted by Arm

Take a look at Souten Kouro. I'm no encoder, but he said it's a good example of a 340mb release not being enough.

If a 340MB release is "not enough" for an episode, why can't people just suck it up and use CRF rather than forcing all episodes, even those with different content, to the same bitrate?

[Arm]

Originally Posted by Dark Shikari

If a 340MB release is "not enough" for an episode, why can't people just suck it up and use CRF rather than forcing all episodes, even those with different content, to the same bitrate?

I agree. Did my post suggest otherwise or were you just saying that in general and quoting me?

[Kristen]

Originally Posted by Waryas

I'm sorry, not to troll or anything but can i ask what kind of filtering you used ?
The banding on the last panel looks pretty bad.
Can you post the source, UNFILTERED ? (just ivtc + resize/crop, nothing else)

so yea, if you're going to oversmooth and use LSF (thus introduce banding), you're gonna get the same shitty result whatever the size because the video already lost details (Gradient in this case).

On the other hand, if you kept the filtering minimal and actually added a debander to your script then the 700MB would look MUCH better.

There's an appropriate time for each size and it depends on your filtering : If you oversmooth, low size will do and this is what happened in this case.

LimitedSharpenFaster(strength=60)
dfttest(sigma=0.5)

http://img23.imageshack.us/img23/2127/source.png

[mieruno]

Originally Posted by Kristen

LimitedSharpenFaster(strength=60)
dfttest(sigma=0.5)

http://img23.imageshack.us/img23/2127/source.png

i honestly hope that's not the order you filtered it

[Waryas]

Originally Posted by Kristen

LimitedSharpenFaster(strength=60)
dfttest(sigma=0.5)

http://img23.imageshack.us/img23/2127/source.png

notice how your png is bigger and how the banding is less apparent on his shirt.
this is the effect of using LSF without the appropriate masks, it also affects noises / existing bandings and make it worse and then you're denoising/smoothing it (and that will make the "sharpened" banding even worse).

For me, filtering isn't about increasing compressibility but making it look better and then choose the adequate size to retain most of the enhancements.
In your next test try encoding the new .avs to both 350 & 700MB and compare the difference again (Well whenever you have free time, lol)

[Kristen]

Originally Posted by Waryas

notice how your png is bigger and how the banding is less apparent on his shirt.
this is the effect using LSF without the appropriate masks, it also affects noises / existing bandings and make it worse and then you're denoising.

You better denoise first and then use LSF with another edge mask than the default one.

In your next test try encoding that .avs to both 350 & 700MB and compare the difference again (Well whenever you have free time, lol)

It's called :effort:

I've been looking into debanding and dotcrawl a lot for K-On's widescreen version, but you have to realize I'm putting like no effort into the first 13 encodes of this series. Someone gave me that chain 4 months ago, and I've been using it because I honestly don't care.

And yes, I do it in a different order usually. I just don't bother changing it here because I don't care. >.>

[Waryas]

What i'm trying to say, next time you want to do a compressibility test with x264 (350mb vs 700mb or whatever)
just try to do it with little to no filtering.

For debanding, there's gradfun2dbmod which is pretty good but you're gonna need way more bitrate to keep it in your final encode. (Try a CRF encode first to estimate the bitrate needed)
As for de-dotcrawl, there's no magical solution, try and see by yourself the filters out there to find whatever you like best.