On bit depth and DR

[BernardLanguillier]

I sometimes hear that DR and smooth transitions are not that important anymore. I have recently come accross interesting shooting conditions that would have challenged to death any of the DSLR I have previously used. I was extremely impressed by the way the D3x dealt with it and I thought it could be useful to share an image:



The critical part is the sky where we have a significantly under-exposed red channel in a very strong gradient but still don't see any hint of banding.

This is by the way a four images panorama in the 60+ mega pixel range.

Cheers,
Bernard

[douglasf13]

I sometimes hear that DR and smooth transitions are not that important anymore. I have recently come accross interesting shooting conditions that would have challenged to death any of the DSLR I have previously used. I was extremely impressed by the way the D3x dealt with it and I thought it could be useful to share an image:



The critical part is the sky where we have a significantly under-exposed red channel in a very strong gradient but still don't see any hint of banding.

This is by the way a four images panorama in the 60+ mega pixel range.

Cheers,
Bernard

 Beautiful shot.

[lattiboy]

Thanks for this good example of bit-depth! I've never gotten a straight explanation to how it would help in non-theoretical terms. Do you notice a big difference from your previous camera? Do you have an example of shot where this would've been extremely helpful?

There are rumors of the A900 implementing this via firmware (I imagine they'll use oversampling?) and it will be great to do some before and after shots to compare.


PS beautiful picture!

[Panopeeper]

I sometimes hear that DR and smooth transitions are not that important anymore. I have recently come accross interesting shooting conditions that would have challenged to death any of the DSLR I have previously used

Bernard,

I am a proponent of 14bit depth with lossless data and participating in discussions about the merit of these aspects. I am "collecting" raw images with banding, which may be attributable to gaps in the levels. However, I accept when the "other side" says that a proof requires two shots: one lossy with banding/posterization and one lossless, without the posterisation. Or one with 12bit depth and posterizing, the other with 14bit depth and not posterizing.

It could have been very interesting to have the same shots both with lossless and lossy data. Btw, increasing the intensity contributes to the banding.

____________________

Another issue: what material is the roof covered with?

[fike]

Bernard,

I am a proponent of 14bit depth with lossless data and participating in discussions about the merit of these aspects. I am "collecting" raw images with banding, which may be attributable to gaps in the levels. However, I accept when the "other side" says that a proof requires two shots: one lossy with banding/posterization and one lossless, without the posterisation. Or one with 12bit depth and posterizing, the other with 14bit depth and not posterizing.

It could have been very interesting to have the same shots both with lossless and lossy data. Btw, increasing the intensity contributes to the banding.

____________________

Another issue: what material is the roof covered with?

Unfortunately, too often the banding is caused by the jpg compression for the web.  Banding seems also to be most prevalent when you are pushing the limits of curves or levels adjustments (compressing the tonal range too much).  Banding directly out of the camera is somewhat more difficult to find.  I find that it is more frequently observable when specific colors clip and create off-color transitions towards the blown highlights of white.  In a deep-blue sky like Bernard shows us, I don't think you could easily get much banding.

[Panopeeper]

Unfortunately, too often the banding is caused by the jpg compression for the web

The lossiness of the JPEG data does not cause any banding; it can cause only local artifacts. The banding is caused by the lack of transitional tone levels, often caused by "stretching" the data (increasing the contrast and/or the brightness).

[ejmartin]

The lossiness of the JPEG data does not cause any banding; it can cause only local artifacts. The banding is caused by the lack of transitional tone levels, often caused by "stretching" the data (increasing the contrast and/or the brightness).

I think he may have been referring to the fact that truncation to 8-bit tonal depth in jpegs from a higher bit depth source is often the cause of banding in web images.  At least that's what I would have said.  

Stretching the tonal levels when the noise is larger than the quantization step does not generate banding.  Compression of the tonal levels can cause the noise to drop below the level spacing, in which case the noise no longer dithers the transitions and banding results.

[lovell]

The title of this thread is "On bit depth and DR, An interesting sample".

But bit depth and DR are not the same thing!

What this thread really is about is tonal graduations, and NOT DR.

[Luis Argerich]

Very nice thread.
I'd love a 12bit vs 14bit example where I can see a difference.

Luigi