encoded size gets bigger with more analyse

[gigah72]

i have a test clip (2nd chapter from PI) and i made a test encode with this setting:
--crf 19 --ref 3 --bframes 4 --b-pyramid --weightb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b8x8,i8x8 --8x8dct --no-fast-pskip --subme 2 --mixed-refs --progress --threads auto --no-psnr --no-ssim --no-dct-decimate --level 4 --cqmfile "prestige.cqm" --aq-strength 0.5 --aq-sensitivity 11

resulting size is 119mb, now i only added i4x4 to analyse and the output file grows to 131mb size. (the source is from dvd and resized via avisynth to 656x384, little Fluxsmooth filtering, movie is extrem dirty&grainy, black/white, shaking camera, etc.)

shouldn't more analyse options increase the compression?

[R3Z]

--subme 2 is your problem. Make it at least --subme 5 otherwise its not worth compressing in my opinion.

[gigah72]

why should --subme 2 be the problem?
with --subme 5 & i4x4 the result is 126mb, beside a drop of ~25% in encoding speed.

i thought it could be due to the source, so i made the same test with an other clip (MATRIX, chapter 1, 15, 29)

with same basic settings it's 141mb, with --subme 5 & i4x4 instead it's 146mb and speed drop.

[R3Z]

why should --subme 2 be the problem?
with --subme 5 & i4x4 the result is 126mb, beside a drop of ~25% in encoding speed.

i thought it could be due to the source, so i made the same test with an other clip (MATRIX, chapter 1, 15, 29)

with same basic settings it's 141mb, with --subme 5 & i4x4 instead it's 146mb and speed drop.

It could be that the sources you tested have a very low bit per pixel in mpeg2 form.

In any case, x264 aint a miricale worker. There is no preset size when using crf because its giving a certain quality, not bitrate. Its one of the caveats of crf. If size matters, use 2 pass bitrate.

Its never a good idea to skimp on subme or macroblock analysis, because they define the quality/compression you get.

Try including trellis 1 and up the subme to 6 and use upto 6 ref frames as well as rdo for bframes. Dont whinge about speed when asking for compression of already compressed sources, as i said x264 isnt a miracle worker.

[Sagittaire]

why should --subme 2 be the problem?
with --subme 5 & i4x4 the result is 126mb, beside a drop of ~25% in encoding speed.

i thought it could be due to the source, so i made the same test with an other clip (MATRIX, chapter 1, 15, 29)

with same basic settings it's 141mb, with --subme 5 & i4x4 instead it's 146mb and speed drop.

Well it's certainely not the same quality. Same size doesn't mean same quality. IMO 126mb with --subme 5 is by far better for quality than 119 mb with --subme 2.

[gigah72]

it's not about difference in size with different --subme, as it's clear that higher subme gives lower filesize, like it's 131 (subme 2) vs 126 (subme 5), both with i4x4 enabled.
but why output gets bigger when i also use i4x4 in the analyse?
how can i meassure the mentioned bit/pixel ratio from my source?

bitrate viewer gives me for the pi sample:
http://img84.imageshack.us/img84/2818/pizc1.png (http://imageshack.us)
Shot at 2007-08-18

and this for the matrix sampe:
http://img225.imageshack.us/img225/5423/matrixfx7.png (http://imageshack.us)
Shot at 2007-08-18


does this explain anything for the particular i4x4 increase?

[Terranigma]

You may find this (http://forum.doom9.org/showpost.php?p=1033570&postcount=24) test I conducted interesting. Check the results for Tests 2 & 3 and maybe do some testing of your own; perhaps an ssim test.

[gigah72]

i reencoded with ssim and psnr output, but i don't know how to read the ssim/psnr results :confused:

1. subme 2 (135mb)
2. subme 6 (129mb)
3. subme 2 + i4x4 (145mb)
4. subme 6 + i4x4 (131mb)

could someone please explain what it means in order of resulting quality? :thanks:



C:\_download\_t_PI>"C:\_sysprg\staxrip\custom\Applications\x264\x264.exe" --crf 20 --ref 3 --bframes 4 --b-pyramid --wei
ghtb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b8x8,i8x8 --8x8dct --no-fast-pskip --subme 2 --m
ixed-refs --progress --threads auto --no-dct-decimate --level 4 --cqmfile "c:\_sysprg\staxrip\matrix\prestige.cqm" --aq-
strength 0.5 --aq-sensitivity 11 --deadzone-inter 9 --deadzone-intra 5 --ipratio 1.1 --pbratio 1.1 --output "C:\_downloa
d\_t_PI\VTS_02_1_EncoderOutput.s2.mp4" "C:\_download\_t_PI\VTS_02_1.avs"
avis [info]: 656x384 @ 25.00 fps (6798 frames)
x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 2 (scale 25)
x264 [info]: slice I:83 Avg QP:23.47 size: 50201 PSNR Mean Y:42.72 U:95.56 V:94.71 Avg:44.44 Global:43.78
x264 [info]: slice P:3793 Avg QP:24.15 size: 25385 PSNR Mean Y:38.50 U:96.18 V:95.65 Avg:40.25 Global:39.35
x264 [info]: slice B:2922 Avg QP:24.54 size: 14048 PSNR Mean Y:38.58 U:95.92 V:95.35 Avg:40.32 Global:39.06
x264 [info]: mb I I16..4: 20.5% 30.9% 48.5%
x264 [info]: mb P I16..4: 15.5% 28.4% 0.0% P16..4: 26.4% 15.1% 13.7% 0.0% 0.0% skip: 0.9%
x264 [info]: mb B I16..4: 2.0% 6.2% 0.0% B16..8: 42.3% 3.5% 11.0% direct:17.6% skip:17.5%
x264 [info]: 8x8 transform intra:64.7% inter:26.0%
x264 [info]: ref P 71.2% 16.9% 11.9%
x264 [info]: ref B 82.5% 11.3% 6.2%
x264 [info]: SSIM Mean Y:0.9635770
x264 [info]: PSNR Mean Y:38.590 U:96.064 V:95.510 Avg:40.330 Global:39.255 kb/s:4162.97

encoded 6798 frames, 19.79 fps, 4163.37 kb/s



C:\_download\_t_PI>"C:\_sysprg\staxrip\custom\Applications\x264\x264.exe" --crf 20 --ref 3 --bframes 4 --b-pyramid --wei
ghtb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b8x8,i8x8 --8x8dct --no-fast-pskip --subme 6 --m
ixed-refs --progress --threads auto --no-dct-decimate --level 4 --cqmfile "c:\_sysprg\staxrip\matrix\prestige.cqm" --aq-
strength 0.5 --aq-sensitivity 11 --deadzone-inter 9 --deadzone-intra 5 --ipratio 1.1 --pbratio 1.1 --output "C:\_downloa
d\_t_PI\VTS_02_1_EncoderOutput.s6.mp4" "C:\_download\_t_PI\VTS_02_1.avs"
avis [info]: 656x384 @ 25.00 fps (6798 frames)
x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 2 (scale 25)
x264 [info]: slice I:83 Avg QP:23.47 size: 43930 PSNR Mean Y:41.75 U:95.56 V:94.71 Avg:43.48 Global:42.81
x264 [info]: slice P:3793 Avg QP:24.15 size: 23581 PSNR Mean Y:38.88 U:96.19 V:95.65 Avg:40.62 Global:39.79
x264 [info]: slice B:2922 Avg QP:24.54 size: 14539 PSNR Mean Y:38.63 U:95.93 V:95.35 Avg:40.37 Global:39.12
x264 [info]: mb I I16..4: 2.9% 90.4% 6.8%
x264 [info]: mb P I16..4: 0.3% 24.6% 0.0% P16..4: 38.6% 21.1% 12.1% 0.0% 0.0% skip: 3.2%
x264 [info]: mb B I16..4: 4.4% 7.2% 0.0% B16..8: 48.8% 4.5% 11.6% direct:10.7% skip:12.9%
x264 [info]: 8x8 transform intra:89.2% inter:43.9%
x264 [info]: ref P 68.5% 18.1% 13.3%
x264 [info]: ref B 82.7% 11.1% 6.1%
x264 [info]: SSIM Mean Y:0.9618464
x264 [info]: PSNR Mean Y:38.811 U:96.067 V:95.511 Avg:40.551 Global:39.513 kb/s:3988.60

encoded 6798 frames, 12.58 fps, 3988.99 kb/s



C:\_download\_t_PI>"C:\_sysprg\staxrip\custom\Applications\x264\x264.exe" --crf 20 --ref 3 --bframes 4 --b-pyramid --wei
ghtb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b8x8,i8x8,i4x4 --8x8dct --no-fast-pskip --subme
6 --mixed-refs --progress --threads auto --no-dct-decimate --level 4 --cqmfile "c:\_sysprg\staxrip\matrix\prestige.cqm"
--aq-strength 0.5 --aq-sensitivity 11 --deadzone-inter 9 --deadzone-intra 5 --ipratio 1.1 --pbratio 1.1 --output "C:\_do
wnload\_t_PI\VTS_02_1_EncoderOutput.s6i.mp4" "C:\_download\_t_PI\VTS_02_1.avs"
avis [info]: 656x384 @ 25.00 fps (6798 frames)
x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 2 (scale 25)
x264 [info]: slice I:83 Avg QP:23.47 size: 43930 PSNR Mean Y:41.75 U:95.56 V:94.71 Avg:43.48 Global:42.81
x264 [info]: slice P:3793 Avg QP:24.15 size: 23917 PSNR Mean Y:38.98 U:96.19 V:95.65 Avg:40.72 Global:39.87
x264 [info]: slice B:2922 Avg QP:24.54 size: 15036 PSNR Mean Y:38.72 U:95.93 V:95.35 Avg:40.46 Global:39.20
x264 [info]: mb I I16..4: 2.9% 90.4% 6.8%
x264 [info]: mb P I16..4: 0.3% 24.6% 1.0% P16..4: 38.3% 20.7% 11.9% 0.0% 0.0% skip: 3.2%
x264 [info]: mb B I16..4: 4.3% 6.2% 2.1% B16..8: 48.5% 4.4% 11.0% direct:10.6% skip:12.9%
x264 [info]: 8x8 transform intra:82.9% inter:45.0%
x264 [info]: ref P 68.5% 18.1% 13.4%
x264 [info]: ref B 82.9% 11.1% 6.0%
x264 [info]: SSIM Mean Y:0.9622925
x264 [info]: PSNR Mean Y:38.904 U:96.068 V:95.512 Avg:40.643 Global:39.593 kb/s:4068.75

encoded 6798 frames, 11.91 fps, 4069.14 kb/s



C:\_download\_t_PI>"C:\_sysprg\staxrip\custom\Applications\x264\x264.exe" --crf 20 --ref 3 --bframes 4 --b-pyramid --wei
ghtb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b8x8,i8x8,i4x4 --8x8dct --no-fast-pskip --subme
2 --mixed-refs --progress --threads auto --no-dct-decimate --level 4 --cqmfile "c:\_sysprg\staxrip\matrix\prestige.cqm"
--aq-strength 0.5 --aq-sensitivity 11 --deadzone-inter 9 --deadzone-intra 5 --ipratio 1.1 --pbratio 1.1 --output "C:\_do
wnload\_t_PI\VTS_02_1_EncoderOutput.s2i.mp4" "C:\_download\_t_PI\VTS_02_1.avs"
avis [info]: 656x384 @ 25.00 fps (6798 frames)
x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 2 (scale 25)
x264 [info]: slice I:83 Avg QP:23.47 size: 50201 PSNR Mean Y:42.72 U:95.56 V:94.71 Avg:44.44 Global:43.78
x264 [info]: slice P:3793 Avg QP:24.15 size: 27864 PSNR Mean Y:39.00 U:96.18 V:95.65 Avg:40.74 Global:39.77
x264 [info]: slice B:2922 Avg QP:24.54 size: 14675 PSNR Mean Y:38.71 U:95.93 V:95.35 Avg:40.45 Global:39.17
x264 [info]: mb I I16..4: 20.5% 30.9% 48.5%
x264 [info]: mb P I16..4: 14.8% 19.5% 13.9% P16..4: 25.7% 13.7% 11.5% 0.0% 0.0% skip: 0.9%
x264 [info]: mb B I16..4: 1.8% 4.7% 2.8% B16..8: 41.8% 3.3% 10.3% direct:17.7% skip:17.5%
x264 [info]: 8x8 transform intra:41.3% inter:25.7%
x264 [info]: ref P 71.6% 16.7% 11.7%
x264 [info]: ref B 82.8% 11.1% 6.1%
x264 [info]: SSIM Mean Y:0.9654593
x264 [info]: PSNR Mean Y:38.921 U:96.065 V:95.511 Avg:40.661 Global:39.531 kb/s:4493.59

encoded 6798 frames, 18.80 fps, 4493.99 kb/s

[Dark Shikari]

1. subme 2 (135mb) 0.9635770
2. subme 6 (129mb) 0.9618464
3. subme 2 + i4x4 (145mb) 0.9622925
4. subme 6 + i4x4 (131mb) 0.9654593

Very strange, --subme 6 actually lowers the SSIM in 2), which is really weird.

You should probably be using --b-rdo. The weirdness could be coming from your CQM also.

[akupenguin]

Yep, prestige.cqm completely screws up intra analysis.

[gigah72]

Yep, prestige.cqm completely screws up intra analysis.

but the resulting output looks best to everything else i tested and compared :cool:

[Dark Shikari]

but the resulting output looks best to everything else i tested and compared :cool:
But its 4000kbps. That's hardly smaller than the original DVD to begin with. Of course it might look better, but then why re-encode to begin with?

[gigah72]

the sample in this test has an average of over 8000kbps, see the 1st screenshot from bitrate viewer. so 4000kbps for 85-90% of the original quality is for me a good result.

[CruNcher]

try crf 25 should give you a noticeable compression gain and is worth the 11 fps, dont forget the higher (lower quant) the bitrate the slower H264 gets @ encoding and the lower the bitrate (higher quant) the faster :) so you would speed up encoding i would go low as a ssim of 0.942xxxx but looking @ the psnr it's quite low allready lower then 40 normaly degrades alot (but ssim looks still ok hmm). Please use crf 25 and tell me the PSNR,SSIM and your Visual Impression of the encode, are you intentionaly blur the source (to get better compression) with deadzone-inter/intra 9/5 ? :)

aku what about fixing x264 on the prestige matrix instead the flat one also RC i think it would give better results then now for all kind of sources HD/SD :)
at qp 1 prestige results ssim wise almost @ the max nearer then with flat or jvt (ofcourse bitrate is also alot higher)

[gigah72]

here's the log from crf20 vs crf25, i did both, 'cause it's on an other pc:


C:\_daten\iso\DVDFab\_t_PI>"C:\sysprg\staxrip\custom\Applications\x264\x264.exe" --crf 20 --ref 3 --
bframes 4 --b-pyramid --weightb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b
8x8,i8x8 --8x8dct --no-fast-pskip --subme 2 --mixed-refs --progress --threads auto --no-dct-decimate
--level 4 --cqmfile "c:\sysprg\staxrip\matrix\prestige.cqm" --aq-strength 0.5 --aq-sensitivity 11 -
-deadzone-inter 9 --deadzone-intra 5 --ipratio 1.1 --pbratio 1.1 --output "C:\_daten\iso\dvdfab\_t_P
I\VTS_01_1_EncoderOutput.crf20.mp4" "C:\_daten\iso\dvdfab\_t_PI\VTS_01_1.avs"
avis [info]: 656x384 @ 25.00 fps (6798 frames)
x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 2 (scale 25)
x264 [info]: slice I:102 Avg QP:23.59 size: 52091 PSNR Mean Y:42.36 U:95.90 V:95.12 Avg:44.08 Gl
obal:43.56
x264 [info]: slice P:3793 Avg QP:24.14 size: 25213 PSNR Mean Y:38.50 U:96.19 V:95.67 Avg:40.25 Gl
obal:39.35
x264 [info]: slice B:2903 Avg QP:24.55 size: 14004 PSNR Mean Y:38.57 U:95.90 V:95.32 Avg:40.31 Gl
obal:39.04
x264 [info]: mb I I16..4: 19.7% 30.8% 49.5%
x264 [info]: mb P I16..4: 15.5% 28.1% 0.0% P16..4: 26.6% 15.2% 13.7% 0.0% 0.0% skip: 1.0%
x264 [info]: mb B I16..4: 2.0% 6.2% 0.0% B16..8: 42.4% 3.4% 10.9% direct:17.6% skip:17.5%
x264 [info]: 8x8 transform intra:64.1% inter:26.0%
x264 [info]: ref P 71.3% 16.9% 11.8%
x264 [info]: ref B 82.4% 11.4% 6.3%
x264 [info]: SSIM Mean Y:0.9635658
x264 [info]: PSNR Mean Y:38.592 U:96.063 V:95.510 Avg:40.333 Global:39.253 kb/s:4165.94

encoded 6798 frames, 6.40 fps, 4166.37 kb/s



C:\_daten\iso\DVDFab\_t_PI>"C:\sysprg\staxrip\custom\Applications\x264\x264.exe" --crf 25 --ref 3 --
bframes 4 --b-pyramid --weightb --b-bias 40 --bime --direct temporal --filter -2:-1 --analyse p8x8,b
8x8,i8x8 --8x8dct --no-fast-pskip --subme 2 --mixed-refs --progress --threads auto --no-dct-decimate
--level 4 --cqmfile "c:\sysprg\staxrip\matrix\prestige.cqm" --aq-strength 0.5 --aq-sensitivity 11 -
-deadzone-inter 9 --deadzone-intra 5 --ipratio 1.1 --pbratio 1.1 --output "C:\_daten\iso\dvdfab\_t_P
I\VTS_01_1_EncoderOutput.crf25.mp4" "C:\_daten\iso\dvdfab\_t_PI\VTS_01_1.avs"
avis [info]: 656x384 @ 25.00 fps (6798 frames)
x264 [info]: using cpu capabilities: MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 2 (scale 25)
x264 [info]: slice I:108 Avg QP:28.64 size: 33054 PSNR Mean Y:38.71 U:96.05 V:95.19 Avg:40.44 Gl
obal:39.94
x264 [info]: slice P:3789 Avg QP:29.13 size: 14397 PSNR Mean Y:36.43 U:96.12 V:95.48 Avg:38.17 Gl
obal:37.26
x264 [info]: slice B:2901 Avg QP:29.52 size: 6899 PSNR Mean Y:36.58 U:95.82 V:95.14 Avg:38.32 Gl
obal:37.02
x264 [info]: mb I I16..4: 24.1% 39.4% 36.5%
x264 [info]: mb P I16..4: 18.3% 23.5% 0.0% P16..4: 36.5% 12.6% 6.4% 0.0% 0.0% skip: 2.6%
x264 [info]: mb B I16..4: 1.6% 4.1% 0.0% B16..8: 40.0% 1.5% 4.2% direct:18.1% skip:30.4%
x264 [info]: 8x8 transform intra:56.7% inter:28.2%
x264 [info]: ref P 72.1% 16.4% 11.5%
x264 [info]: ref B 81.3% 11.9% 6.7%
x264 [info]: SSIM Mean Y:0.9441952
x264 [info]: PSNR Mean Y:36.530 U:95.995 V:95.334 Avg:38.271 Global:37.186 kb/s:2298.77

encoded 6798 frames, 6.71 fps, 2299.21 kb/s


@ crf25 it's really ugly, blocks everywhere, something like pumping, but this could be due to the crap display i have here in this place.

why does lowering the deadzone settings imply that the image gets blured?
from what i read about it, lowering values means to let pass more noise, grain, etc. and this is what i want to preserve, to a certain degree.

[akupenguin]

aku what about fixing x264 on the prestige matrix instead the flat one
Maybe it's theoretically possible to do something sane with prestige, but I certainly don't know how. I can't even attempt to handle matrices that quantize some frequencies 28 times more than others.

at qp 1 prestige results ssim wise almost @ the max nearer then with flat or jvt (ofcourse bitrate is also alot higher)
Higher bitrate than lossless, you mean?

[check]

Yep, prestige.cqm completely screws up intra analysis.
What exactly is the practical upshot in terms of x264 decisions that the prestige matrix changes? Also, have you ever made any matrices for AVC worth sharing?

[akupenguin]

If you enable RDO, many macroblocks that really should be intra (because they're unrelated to the previous frame) get encoded as inter with a random mv (because prestige's intra cqms have too many small coefficients, intra takes more bits than appropriate for a given qp).
x264 treats partition decision and dct size decision as orthogonal, but prestige's 4x4 and 8x8 inter cqms are significantly different, so the partition that's optimal for one dct might not be optimal for the other.
If you quantize some frequencies 28 times more than others, but give them equal SSD weights, that sets up a conflict whereby RDO tries to put bits back into the high frequencies any way it can.

If you disable RDO, you get the opposite symptoms of the same problem: It doesn't avoid intra (i.e. decides under the assumption that all of the matrices are the same), and thus wastes bits kinda at random.

It might be possible to tune a codec for a specific matrix. But the only way I know of to fix the above problems in general is to use RD-AQ for each partition for each dct size. This would increase the number of RD evaluations by maybe a factor of 5 or 10, and reduce speed commensurately.
You would also have to use a distortion metric appropriate to your matrix. But you can't just weight the SSD the same way as the cqm, because there are 4 different matrices, so the can't all be correct.

[Terranigma]

Yep, prestige.cqm completely screws up intra analysis.
but the resulting output looks best to everything else i tested and compared :cool:

I have the agree with akupenguin here; it totally screws up the analysis. I switched from the default Quantizer to "Prestige" only because i've noticed the majority of the folks around here were using it. Maybe until now, because I did a test, and the result's probably shocking, but not shocking to akupenguin

Test 1 (Prestige CQM):
x264667b --bitrate 1301 --keyint 125 --min-keyint 12 --ref 5 --mixed-refs --no-fast-pskip --bframes 3 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -3,-3 --subme 7 --trellis 2 --partitions p8x8,b8x8,i4x4,i8x8 --8x8dct --b-bias 40 --me dia --threads auto --thread-input --cqmfile "Prestige CQM.cfg" --progress --no-dct-decimate --aq-strength 0.3 --aq-sensitivity 5.0 --output "C:\MytestClip.mkv"

Test 2 (Default Quantizer):
x264667b --bitrate 1301 --keyint 125 --min-keyint 12 --ref 5 --mixed-refs --no-fast-pskip --bframes 3 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -3,-3 --subme 7 --trellis 2 --partitions p8x8,b8x8,i4x4,i8x8 --8x8dct --b-bias 40 --me dia --threads auto --thread-input --progress --no-dct-decimate --aq-strength 0.3 --aq-sensitivity 5.0 --output "C:\MytestClip.mkv"



Test 1 Statistics:
SSIM Mean Y:0.9740672
PSNR Mean Y:42.008 U:47.035 V:47.422 Avg:43.125 Global:42.159
(encoded frames: 3000 @ 3.78fps, 1340.42 kb/s)
(final ratefactor: 21.85)
(8x8 transform intra:87.4% inter:86.6%)

Test 2 Statistics:
SSIM Mean Y:0.9746506
PSNR Mean Y:42.357 U:47.256 V:47.655 Avg:43.461 Global:42.527
(encoded frames: 3000 @ 4.03fps, 1336.36 kb/s)
(final ratefactor: 21.36)
(8x8 transform intra:67.2% inter:80.6%)

With the default quantizer, I got a faster encoding speed as well. :D

---
forgot to add that the resolution was of the same exact size gigah72 used above.
656x384

[gigah72]

but how can i get the same quality result, without useing this cqm and deadzone?

btw, if someone's interested, here's a short scene (~500 frames, 22mb):
http://rapidshare.com/files/50578715/VTS_01_1_cut1.7z.html

[Terranigma]

Well your source exhibits lots of motion and has a rather high amount of film grain. Grain + motion = bitrate miser, so your source is going to be large regardless; well, unless you like blocky, non-detailed encodes.