Compressibility Test

[CiNcH]

Compressibility Test in Gordian Knot 0.28.8

with DivX Pro 5.1.1 at constant quality Q=2

Source: 1145 frame sample from The Mummy Returns


Compressibility Test 1:
no cropping -> unresized output 720x576

GKnot returns 0,496 bits/pixel


Compressibility Test 2:
cropped top/bottom 74/70 -> resized output 720x304 (Sharp Bicubic)

GKnot returns 0,732 bits/pixel


Conclusion:
It's not possible that the unresized clip requires less bits/pixel than the cropped downsampled clip to be encoded at Q=2.

I have also done the filesize test at Q=2. Unresized clip is 30,3 MB big and the cropped downsampled clip is 23,8 MB big.


Same results with XviD 1.0 RC4!

[manono]

Nice try. All I see that you've proven is that black bars take way fewer bits to encode than the active video.

[jggimi]

It's also a misunderstanding of the formula. The formula is "value = bits / (pixels * frames)"Example #1: 30,3 MB (254.174.822 bits). 414.720 pixels per frame. 1.145 frames. Value = 0,535

Example #2: 23,8 MB (199.648.870 bits). 218.880 pixels per frame. 1.145 frames. Value = 0,796

[CiNcH]

It's also a misunderstanding of the formula.

Only ;) . Thanks for clearing it up!


JUST A BUG IN MY BRAIN.

[CiNcH]

Seems that I have got even more bugs in my brain. Probably another easy answer for me? ;)

DivX 5 Compressibility Check at Q=2

Value in Standard P/Q 0,657
Value in Slowest P/Q 0,689

Meaning that Slowest P/Q mode requires more bits/pixel to encode the sample.

According to this Compressibility Check what can we expect from the Quantizer when we encode the same sample at a given bitrate (tested with 800 kbps). There should be higher Quantization in the Slowest P/Q mode encode, right?

Here are the average Quantizers after encoding the sample with 800 kbps:

Average Q in Standard P/Q 7,428821
Average Q in Slowest P/Q 7,043660

So lower Quantization in the Slowest P/Q mode encode although it requires more bits?


Something's wrong here, isn't it?