DxOMark still silent on the E-M5

Started Sep 1, 2012 | Discussions thread
Anders W Forum Pro • Posts: 21,468
Re: Dxo is a questionable "organization" and . . .

3DrJ wrote:

Anders W wrote:

The only processing required in taking the step from the charge stored in each pixel on the sensor to the RAW output file is to convert from an analog signal of a certain magnitude (an electric charge) to a digital signal of corresponding magnitude (a number). Please tell me how you arrive at the conclusion that this step must necessarily involve some kind of cheating.

Of course it's not really "cheating", but that's been a term others have applied. All I was saying was that of necessity some processing of the sensor's signals will occur. Exactly how it's done may not be revealed to us, wherein some "corrections" may be applied, but still, unless there is spurious camera output, no way to know if some "magic" has occurred. No doubt best to make no assumptions.

The point is that there are ways to reveal whether "magic" has occurred in the sense of "cooking" or "cheating".

Anyone can measure sensor output in the form of a RAW file. As indicated above, the conversion from analog signal to RAW digital data need not (and typically does not) involve more than a per-pixel conversion from an analog signal to a digital one.

I don't know that we can say that for sure. Who knows, at the silicon level that "per-pixel conversion" could be modified or more complex that theory would suggest.

Because there are ways to test statistically for that.

Now lets doff our chef hats to the camera vendors who "cheat". The more they are willing and able to do so, the better the results we get to enjoy.

I am not particularly fond of camera manufacturers who do "cheat" and thus find no reason to "doff my chef hat" to them. AFAIK, "cheating" in the stage from analog to RAW digital output, when it occurs, takes the form of noise reduction, i.e., resolution is traded for lower noise. DxOMark regularly tests for this kind of "cheating" and reports it when they observe it. There may have been cameras who slipped by this control in the past, but I think it unlikely that they would do so at this stage.

Outside of a few at Sony or Olympus, no one can say what "tricks" might be employed to optimize output parameters, such as noise, etc. And if it's the case, we can't describe it as cheating, just process improvements.

We can draw the line between "processing improvements" and "cheating/cooking" and that suffices for me. For example, the manufacturers make strong efforts to reduce read noise (a kind of electronic noise added in the step between electric charge and a digital data point corresponding to that charge). This is not "cheating" or "cooking". On the contrary, it makes the raw material cleaner by reducing a disturbance.

Noise reduction by means of averaging pixel values, on the other hand, just covers up one problem (noise) at the expense of another (loss of resolution). I don't like this being done by the camera since a) I prefer to be in control of when and whether to apply NR, b) I would like to choose the tools myself, c) the tools are likely to become better over time so that I can go back and reprocess my RAWs at a later stage if I care too.

What we can tell are the characteristics of the output. If resolution is only so much under particular conditions, that's a fact. We don't have to account for the how it got to be the way it is, all we have to do is see what it is.

Sites like DxO don't try to measure actual sensor resolution. They only report the pixel count. Therefore, I am grateful that they report whether a manufacturer applies NR in RAW, thus a) demonstrating that the sensor at certain ISOs has less resolution than you might expect and b) accounting for that fact. If they didn't do this, their data on other things (SNR, DR, etc) would be misleading in the comparison between different cameras.

As you say, actually "cheating", or trying to, is not likely to pass unnoticed by sharp observers. Manufacturers who attempt to cheat are likely to lose out rather quickly.

Camera makers won't fool Mother Nature.

Noone can fool Mother Nature. Fortunately, they are not likely, at this stage, to fool DxOMark either.

Here is further information on how DxOMark handles the issue:

Cameras known to do NR in RAW include the Pentax K-5 (from ISO 3200 on), the Nikon series 1 cameras (from ISO 800 on), and the Leica M9 (from ISO 160 on). As you can see if you go to for example the DR graph for the K-5 via the link below, this is indicated by DxOMark by using a "hollow" rather than solid dot for the "smoothed" data.

No doubt there could be many ways output could be less than ideal. What can be done is to document the result, the facts are what they are. We don't have to "explain" it, in fact, it may not even be possible to do so. If "smoothed" output is the best, or all, a camera can do, it's a characteristic to document. Ultimately, it's users who determine the camera's acceptability for their particular purposes.

As to explaining it, see above.

 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +28 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow