Many thanks guys for significant arguments! It may be the quantity of English required for one week for me to understand. LOL

@cbrandin

"I would hope that differences between FM2 and GOLGOP3 looks would be clearly visible - the very purpose of having different matrices is to offer such differences, otherwise why bother? There is no such thing as a perfect matrix - there are ones for particular applications, and others for more general use."

Yes, we must not forget this. Although I do not deny evaluation subjective, of course...

Although I am making various settings, what I aim at is a setting in which a problem does not occur at all in at least 95 MB/sSDXC card use. About subjective liking, I leave it to those who use. And I also think it significant to exchange PERSONAL VIEWs mutually. :-)

BTW, since the confusion about my settings has arisen here, I explain correctly. It is about my settings among three sorts of settings which @towi compared in HBR.

Valkyrie 444 TYPE-ZERO2 C1.3 : With Modified Chris's pure original 444 matrix

GOLGOP3-13-A : With My original matrix named 'EX Matrix'

It is only Valkyrie that I have applied 444 concepts by Chris and Nick. There is character near the Sedna matrix of Nick in 'EX matrix'.

@jrd When you consider that hand-tuned quantization matrices lie at the heart of all JPEG, MJPEG, and H.264 encoders, it's remarkable how little documentation and statistical evaluation I was able to find on their design and performance. In the field of perceptual audio compression, the industry has developed standardized MUSHRA tests to quantitatively evaluate the perceived quality of lossy audio encoding. This practice has had a major impact on the development of new audio compression techniques.

In the world of video, I haven't found any evidence of comparable evaluation tests on video compression quality. I suspect one of the reasons is that video codec development has been predominantly driven by commercial requirements to minimize the data size of video files while maintaining "acceptable" motion picture quality. The conundrum there is that what's considered "acceptable" video quality has changed dramatically in the transition from DVD to Blu-ray media. Moreover, the technical requirements for recording high-quality source video are vastly different from what's needed for consumer-grade distribution of finished products. AVCHD, in particular, was not originally developed or evaluated for use as a motion capture codec.

In short, this entire field is currently a "black art" lacking in well-documented perceptual encoding theory and practice. What exists is engineering lore based on reference implementations from standards committees and track records of encoders successfully used for similar purposes. That makes it an interesting challenge for me, since it requires engineering intuition as well as skillful testing procedures. But to evaluate the results rigorously would require extensive statistical testing programs involving a wide variety of subject matter and human participants. That would be a project whose scope is well outside my own resources.

@cbrandin @Lpowell

Agreed, that differences may not be in the realm of better or worse, and with the additional proviso that gradeability of the footage may render any perceived differences out of the camera moot.

It's just that the differences themselves -- good, bad, agnostic -- remain largely undocumented. Granted, nobody here is getting paid to do that work, and few are well equipped to do it anyway, with controlled conditions. Just that, in the absence of controlled testing, drawing any conclusion can be a challenge, even in the realm of the wholly subjective....

@jrd There are several things to bear in mind with these types of comparisons. If you take a close look at the comparison @towi uploaded in the first post here:

http://www.personal-view.com./talks/discussion/2960/an-valkyrie-golgop3-13-gop3zilla-natural-super-reliable-all-round-gop3-settings-/p26

This video was intentionally shot defocused and underexposed with virtually no edge details. Regardless of the patch, this type of subject will show the GH2 at its very worst. What this reveals is how crudely the camera performs in extremely poor shooting conditions. That's not a criticism of the test - I think it was well chosen to highlight this problematic aspect of the GH2's encoding behavior.

The accompanying Stream Parser chart of the three tested patches (decoder_comp.png) demonstrates another aspect of the encoder's behavior. It shows what each patch does when there's virtually nothing but low-level noise to encode. The numbers in the DC column indicate the minimum degree of quantization used by the encoder in each frame. The lower the number, the finer the quantization, and the higher the required bitrate.

What I can tell you about Flow Motion v2 is that in this case it's performing exactly as intended, encoding all frames at its finest DC quantization level of 5. One reason I chose that number is because it's the finest degree of quantization I was able to persuade the encoder to maintain consistently all the way up to 100Mbps. The other reason, as Chris Brandin has explained, is that DC quantization levels less than 5 are below the mathematical limit of measurable improvements in image quality.

However, that doesn't necessarily mean that extremely fine quantization, such as what GOLGOP3 produces here, has no visible effect on the encoded image. It just means that we can no longer objectively measure its effect on image quality. Likewise, the visual differences between the 444-soft matrix used in this version of GOLGOP3 and the custom matrix in FM2 become increasingly subjective as the content becomes increasingly random.

Personally, after analyzing the technical aspects of the GH2's performance in low-contrast, under-exposed conditions, my conclusion is this: it sucks. The image sensor is too noisy and the encoder has nowhere near enough color depth to do a presentable job of encoding low-level noise. To my eyes, the chroma noise in near black shadows looks especially bad - persistently oversaturated and blotchy with gross macroblock artifacts where it cuts off to pure black. In my view the AVCHD encoder deals with this in a dysfunctional manner, and fails to reduce the saturation of near-black colors according to well-known psycho-visual principles (namely, that human vision becomes increasingly monotonic as illumination is decreased).

Philosophically, I understand that what I consider random noise is what others seek to cultivate as "film grain". I'm not claiming you're wrong, just that esthetically, I don't perceive it that way. If I want to see film grain in a particular video, there are many excellent filters I can use to simulate it in a controlled manner. To facilitate this approach, I tweaked the quantizing matrices I designed for Flow Motion v2 to suppress finely detailed chroma noise, while increasing its sensitivity to fine luma details. The 444-soft matrix used in GOLGOP3 actually does the opposite and the difference is clearly visible to my eyes.

The settings are available for everyone here on the forum. Load them, compare the results and draw your own conclusions. It's not a big deal ...

Consider that there are thousands of posts here, and almost no objective examination of the claims made for any given setting. On the other hand, no setting is going to overcome the limitations of the camera, so the margin of improvement we're arguing over is going to be small; the GH2's own menu settings will likely have a far greater effect. There is, however, a wish that claims would be substantiated.

@jrd I wouldn't mind seeing a comparison either....as long as it's reasonable and doesn't get into the sub atomic level. I kind of hate to see these comparisons get to "my setting is better than yours", though.

bkmcwd, LPowell, Ralph. Driftwood, Vitaliy and a host of others have dedicated a lot of time to come up with these settings and really....made history.

All the settings have their merits and their strong and weak points.

@ GOP3ZILLA lovers :-)

"GOP3ZILLA_3_Candidate 1"

I release the first candidate for "GOP3ZILLA_3" setting.

This is the simple setting which harnessed the feature of the stock as it was.

I think that 1080i and HBR especially in PAL of these settings are improved rather than Previous BETA versions.

Although I am preparing other candidates, if they are not satisfactory, I will make this C1 the official setting.

Please try it! And many thanks for your feedbacks!

*Of course, already combined with excellent audio tune @thepalalias's "Pasadena_Pulse_Audio_V2_Beta2"

NOTE:

*Although bit rate setups are quite high, the bit rate which actually appears are lower than these. Since the actual bit rate changes by the object to shoot, please confirm by yourself.(Max bitrate : about 100M in 24p, about 75M in HBR.)

*24H and 24L are high quality 3GOP setting with mega size I-frames.

*24p80%(30p) is available in 24H & 24L both.

*60i, 50i, HBR30p and HBR25p are also high quality and stable 3GOP settings.

*720/50p and 720/60p are 6GOP and 9GOP. It has tuned up so that the bit rate at the time of motion may go up rather than the time of statics.

*I especially recommend you use with SanDisk 95MB/s 64GB SDXC card.

*In camera playback is available in all rec mode, at least available after reboot the camera.(Not tested full yet, but probably)

*EX Tele, iA mode work.(Not tested full yet, but probably)

*Spanning in 24H, 24L with SanDisk 95MB/s 64GB SDXC Card. In other rec mode, need test.

*Spanning in other cards needs to be tested.

@Mirrorkisser IMO, top priority is given to gradation in Valkyrie. In GOLGOP3-13, I took the balance of gradation and resolving into consideration. Although I am not sure whether it has done just like that... :-)

@ GOLGOP3-13 lovers :-)

"GOLGOP3-13-A_Candidate 1" and "GOLGOP3-13-MAX_Candidate 1"

I release the each first candidates for "GOLGOP3-13-A" and "GOLGOP3-13-MAX" settings.

I think that 1080i and HBR especially in PAL of these settings are improved rather than Previous BETA versions.

Although I am preparing other candidates, if they are not satisfactory, I will make these C1s the official settings.

Please try them! And many thanks for your feedbacks!

Special thanks to @towi! :-)

*Of course, already combined with excellent audio tune @thepalalias's "Pasadena_Pulse_Audio_V2_Beta2"

NOTE:

*Although bit rate setups are quite high, the bit rate which actually appears are lower than these. Since the actual bit rate changes by the object to shoot, please confirm by yourself.(Max bitrate : about 100M in 24p, about 75M in HBR.)

*These two settings are completely the same except 24p.

*24H and 24L are high quality 3GOP setting with mega size I-frames in GOLGOP3-13-A.

*24H and 24L are high quality 3GOP setting with super mega size I-frames in GOLGOP3-13-MAX.

*24p80%(30p) is available in 24H & 24L both.

*60i, 50i, HBR30p and HBR25p are also high quality and stable 3GOP settings.

*720/50p and 720/60p are 6GOP and 9GOP. It has tuned up so that the bit rate at the time of motion may go up rather than the time of statics.

*I especially recommend you use with SanDisk 95MB/s 64GB SDXC card.

*In camera playback is available in all rec mode except 24p in GOLGOP3-13-MAX, at least available after reboot the camera.(Not tested full yet, but probably)

*EX Tele, iA mode work.(Not tested full yet, but probably)

*Spanning in 24H, 24L with SanDisk 95MB/s 64GB SDXC Card. In other rec mode, need test.

*Spanning in other cards needs to be tested.

@Bonzai "AWB" 5 'clicks' toward the blue?
Yes! That's right.Sorry,I forgot to write"B".

Valkyrie 444 TYPE-ZERO2 C1.3 sample
pana 12-35mm F2.8
CONTAX Planar 50mm F1.7
NOS 0,0,0,0 AWB B+5
no color grading

@radikalfilm

Thanks for testing my setting! :-) I am sorry that my explanation is insufficient since I am poor at English. However, since favorite people explain my settings instead of me, it is supplemented.