Jump to content

M8-why 10MB-vs-DMR 20MB


gogopix

Recommended Posts

Advertisement (gone after registration)

On the DPreviw forum there was a thread by Andrej Kolev who is trying to make a DNG raw converter.He may have broken the code on why the M8 has only 10MB files and the DMR 20MB.

 

He says the M8 files have only 8 bits per channel rather than 16, but that they are non linear mapping to 14 bit luminance range per color RGB. That means in fact M8 has less information than DMR - since the DMR is linear and 16 bit.

When you map 8 bits non-linear to 14-bits linear you can recover dynamic range but you do lose detail,since you dont know all the gradations in between. I wonder why Leica would do this.

 

It could be he is wrong, and the lossless compression just codes redundancy and thus all detail can be recovered (just mapping 8 bit to 14 bits will NOT do it.

There are a lot of situations where non linear mappings lose very little since the eye (as well as ear) are logrithmic devices, that is, it can take a factor of 10 in sound to have a perceived 'doubling' of sound (that is why 10-20 db is about 'twice' as loud perceived, but sound energy is actually 10 times!

 

maybe Leica tested and found a clever mapping that preserves detail, but going from 20MB per image to 10 MB for same number of pixels is half the infrormation. Whether it makes any difference only time will tell. If the coding uses redundacy(like patches of same color) and just compresse more cleverly, well then nothing will be lost.

 

But since we have SO much time on our hands I thought it would be interesting to speculate on something besides lenses and Seals?

 

:-)

 

Victor

 

PS pre-order FIRM in place :-)

Link to post
Share on other sites

  • Replies 214
  • Created
  • Last Reply

You don't have to jump through hoops to reduce filesize because the DNG format has built-in support for lossless compression.

 

I think it's simply a case of Leica not enabling it for the DMR and enabling it for the M8.

 

The saving can be significant. Raw Olympus ORF files out of my crappy E-500 are @ 13mb each, but run them through Adobe DNG converter and they shrink to @ 7.5mb, without image degradation or data loss.

 

QED :?)

Link to post
Share on other sites

Oh dear. He may be right but I certainly hope he is wrong.

 

I had assumed the DNGs were being compressed losslessly to account for the difference between the M8 and the DMR. It's interesting, though, that the manual is quite clear that you can store a fixed 93 DNGs per Gb whereas the size of the compressed JPGs can vary. That implies that the DNGs have a fixed size which does indeed suggest they are not being compressed in a winzip sort of way. In a Nikon D2x, you can switch RAW compression on, it is lossless but the file size varies.

 

It 's possible that one of the major design constraints in the M8 was battery life; the more processing you do and the faster you want to do it, the more power you will use and the lower the shot count per charge. If you use an R-D1 with a nearly flat battery, it's clear there's a big increase in consumption of what little capacity remains when the camera is writing to the card and the battery meter dips.

 

Compressing the data in some way therefore helps battery capacity (as well as card capacity) and they may have found the compute power required to do the lossless compression negated the benefit or simply took too long compared to the data transform. All to do with the balance of power consumption and the speed of the DSP in the camera.

 

I can't see why they might have done this other than to boost the battery life number.

Link to post
Share on other sites

Hasselblad/Imacon images from the 22meg CF backs originally recorded dng files directly. Each file was 42.8meg. Then they changed the back's firmware for more speed, smaller file sizes and commonality with the 39meg backs. Now, the images record in 3fr format and Flexcolor gives the option of converting to dng's as they are imported. The imported dng file is 24.5 meg. Either dng opens to the same file size and there doesn't appear to be a difference in a 16bit tif created from either file.

 

So, 24.5meg files from a 22meg 16bit MFDB is about the same ratio as the files created by the M8.

 

Robbe Gibson

Link to post
Share on other sites

Guest chris_h
There are a lot of situations where non linear mappings lose very little since the eye (as well as ear) are logrithmic devices, that is, it can take a factor of 10 in sound to have a perceived 'doubling' of sound (that is why 10-20 db is about 'twice' as loud perceived, but sound energy is actually 10 times!

 

That's right, what you tell about logarithmic devices - but 10 to 20 db is about 3.3 times louder ... 3 is the factor of doubling.

Link to post
Share on other sites

Advertisement (gone after registration)

If the camera is using a logarithmic data compression scheme ( often used in u-law and a-law voice CODECs ), would that not normally mean the spacing between data values is smaller at low ( shadow ) values and larger at high ( highlight ) values.

 

If curves are subsequently used in photoshop to expand highlight information, e.g. add drama to clouds or detail to snow, will there not be a lot of missing levels ?

Alternatively if the log scheme is like a film 'S' curve will the mid tone suffer?

 

It sounds like a bad move not to offer 3 DNG options: lossy, loss-less and uncompressed, and let the user choose the speed/storage balance.

Link to post
Share on other sites

I relly hope they are not compromising image quality just to chase some arbitrary standard of shots per Gb or shorts per battery charge.

 

If true, this is a far more serious compromise than any amount of inconvenience in the operating firmware.

Link to post
Share on other sites

On the DPreviw forum there was a thread by Andrej Kolev who is trying to make a DNG raw converter.He may have broken the code on why the M8 has only 10MB files and the DMR 20MB.

 

He says the M8 files have only 8 bits per channel rather than 16, but that they are non linear mapping to 14 bit luminance range per color RGB. That means in fact M8 has less information than DMR - since the DMR is linear and 16 bit.

 

I have just examined the tag fields of an M8 DNG:

 

NewSubFileType: Main Image

ImageWidth: 3920

ImageLength: 2638

BitsPerSample: 8

Compression: Uncompressed

PhotometricInterpretation: CFA

 

It appears that, indeed, it is uncompressed DNG, and achieves its small size with 8 bits per sample. (I can't comment on the other aspects of Andrej Kolev's statement).

Link to post
Share on other sites

Maybe this is something that they will change in the final firmware (they'd better get a move on if people are to receive their cameras on 1st November), but if the M8 only takes 8 bit images, it will be a farce, IMHO.

Link to post
Share on other sites

Maybe this is something that they will change in the final firmware (they'd better get a move on if people are to receive their cameras on 1st November), but if the M8 only takes 8 bit images, it will be a farce, IMHO.

 

I should have provided more information:

 

Make: "Leica Camera AG"

Model: "M8 Digital Camera"

Software: "1.04"

DNGVersion: 1.0.0.0

UniqueCameraModel: "M8 Digital Camera"

 

Is "Software" the clue about the firmware version?

Link to post
Share on other sites

Well

 

good info, esp the last. if storage is 8 bit then there must be a fixed algorithm. You can, as pointed above get 16 bit DYNAMIC range by 'spreading' the brighter values.

 

On the audio analogy, yes, acoustic POWER is double (3db is what means doubling the power what is basicall log (base 10) .3 = 2

 

however, what I meant was that we PERCEIVE a doubling not at 3db but at 10db, but that is for the psychologists!

 

Good speculations,

 

any mor?

of course, it may be possible to change in firmware to read all 16

 

now, if the 8 bit are really a 'curve fit' to an original 16 bit linear, then the 'loss could be quite a bit lower, maybe negligible

Link to post
Share on other sites

This is interesting and a little worrisome in terms of image quality.

 

But, reading between the lines of what Sean has said elsewhere, I think he is very impressed with the overall image quality, and that it is at least as good, but probably better than the DMR.

 

This is a great use of "all this time on our hands." Better than buying a 1.4 lens before I know whether I'll like to use 35mm or 50 more with the M 8 crop factor.

 

Cheers,

 

Mitchell

Link to post
Share on other sites

The RAW files I shot on my SD card at the demo last Saturday were 20MB files. The files are on my home computer and I'm at the office, so can't provide any more info -- but I am quite certain of that number.

 

Larry

Link to post
Share on other sites

The RAW files I shot on my SD card at the demo last Saturday were 20MB files. The files are on my home computer and I'm at the office, so can't provide any more info -- but I am quite certain of that number.

 

The Leica manual specfies the RAW files as 10.31MB DNGs and the few examples floating around the internet seem to be consistent with that size.

Link to post
Share on other sites

The Leica manual specfies the RAW files as 10.31MB DNGs and the few examples floating around the internet seem to be consistent with that size.

 

 

Ian,

 

I'm aware of that, but the M8 I used produced 20MB files. I wouldn't draw any conclusions from this other than there must be more than one firmware version floating around at the moment.

 

Larry

Link to post
Share on other sites

the M8 I used produced 20MB files. I wouldn't draw any conclusions from this other than there must be more than one firmware version floating around at the moment.

 

If I can get hold of one of those 20 MB DNGs, I'll post back the details of the TIFF fields, including the Software/firmware number.

Link to post
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...