EOS-1D X Mark II Claims of 15 Stops of DR [CR3]

[rfdesigner]

Well there ya go. I guess it could use some kind of non-linear amplification pre-ADC, and compress more than 14 stops into the 14-bit RAW, but I also kind of doubt that is the case as well. Still, would be interesting to see if the camera actually scores over 13 stops engineering DR.

They could do a non-linear compression on the analog, but once you are 14 bit digital you can not go back to 16 bit without loosing resolution. (it is blazingly fast with a look-up table). When you un-compress you loose resolution. That's why I don't believe anyone who claims more stops of DR than bits of data. If they really had the resolution, they would bump up the number of bits.... after all, that's why we went from 12 bit RAW to 14 bit

It depends. There is a massive amount of highlight information, more than you really need. If you compress properly, what you lose isn't necessarily as important as what you gain. It's 14 bits of data, but that data could represent more than 14 stops of information.

agreed! The highlight information is nowhere near as critical as the lows....going from 11 to 13 while skipping 12 is significant, while going from 16371 to 16379 may be 4 times the jump, but the difference in an image would be almost completely undetectable.

Exactly! Hah! Glad someone gets it.

Yes, but I still think if they had that much DR they would go 16 bit....the marketing people would go nuts "Our new 1DX2 has so much dynamic range that we had to expand our RAW files to 16 bits to handle it!!!! Those inferior Sony and Nikon cameras only need 14 bits with their inferior colour depth...."

I dunno. I don't think the two are synomous. There is having more true dynamic range, which in the case of a linear sensor means you have lower read noise and thus can actually benefit from 16-bit data. A non-linear sensor, on the other hand, amplifies shallow signals more and changes the relationship of those shallow signals vs. read noise more than it does deeper signals. If Canon has not reduced their read noise enough to actually support 16 stops of linear dynamic range, then going to 16-bit wouldn't help. However a non-linear sensor capable of 15 stops of DR compressed into 14-bit data WOULD help.

I would obviously prefer to have a linear sensor with read noise low enough to support 15 stops of DR. I think that would be much better than applying curves to the information to compress and decompress it. A non-linear sensor would just be a means of overcoming high read noise...it would just be a stop-gap measure. However Canon would have to get their read noise down to around 2e- with a 60ke- FWC (i.e. 5D IV), or 3e- with a 100ke- FWC (i.e. 1D X II), in order to have 15 true stops of linear DR. Rather doubtful they have achieved that...I might believe they have reduced read noise down to the ~10e- range @ ISO 100...that would be almost 1/4 what the 1D X had, and if the II still has around 90ke- FWC, then that would be ~13.2 stops DR (I'd take that, though!!). Sony has barely achieved that even...most FF exmors have 4-6e- read noise at ISO 100, and the A7s has ~25e- (although it also has 150ke- FWC).

RE pre-de-emphasis.. really what is the point?.. what is easier to encode, 14 bits non-linear or 16 bits linear, CR2 files already do lossless compression of 14 bits into about 9 on average, card costs are now almost zero compared to camera costs. If it were my decision I'd go with CR3 which would merely be 16bit data with TIFF compression and keep everything linear. Perhaps Canon see CR3 as being viewed as a potential problem so want everyone on side (Adobe etc) before letting that little nugget of info out (after all Pros want things to "just work" and don't care so much about specmanship).. but that's pure speculation.

It would explain them only releasing jpegs

 

[kaihp]

It depends. There is a massive amount of highlight information, more than you really need. If you compress properly, what you lose isn't necessarily as important as what you gain. It's 14 bits of data, but that data could represent more than 14 stops of information.

agreed! The highlight information is nowhere near as critical as the lows....going from 11 to 13 while skipping 12 is significant, while going from 16371 to 16379 may be 4 times the jump, but the difference in an image would be almost completely undetectable.

Exactly! Hah! Glad someone gets it.

Think "floating point numbers"

 

[jrista]

Well there ya go. I guess it could use some kind of non-linear amplification pre-ADC, and compress more than 14 stops into the 14-bit RAW, but I also kind of doubt that is the case as well. Still, would be interesting to see if the camera actually scores over 13 stops engineering DR.

They could do a non-linear compression on the analog, but once you are 14 bit digital you can not go back to 16 bit without loosing resolution. (it is blazingly fast with a look-up table). When you un-compress you loose resolution. That's why I don't believe anyone who claims more stops of DR than bits of data. If they really had the resolution, they would bump up the number of bits.... after all, that's why we went from 12 bit RAW to 14 bit

It depends. There is a massive amount of highlight information, more than you really need. If you compress properly, what you lose isn't necessarily as important as what you gain. It's 14 bits of data, but that data could represent more than 14 stops of information.

agreed! The highlight information is nowhere near as critical as the lows....going from 11 to 13 while skipping 12 is significant, while going from 16371 to 16379 may be 4 times the jump, but the difference in an image would be almost completely undetectable.

Exactly! Hah! Glad someone gets it.

Yes, but I still think if they had that much DR they would go 16 bit....the marketing people would go nuts "Our new 1DX2 has so much dynamic range that we had to expand our RAW files to 16 bits to handle it!!!! Those inferior Sony and Nikon cameras only need 14 bits with their inferior colour depth...."

I dunno. I don't think the two are synomous. There is having more true dynamic range, which in the case of a linear sensor means you have lower read noise and thus can actually benefit from 16-bit data. A non-linear sensor, on the other hand, amplifies shallow signals more and changes the relationship of those shallow signals vs. read noise more than it does deeper signals. If Canon has not reduced their read noise enough to actually support 16 stops of linear dynamic range, then going to 16-bit wouldn't help. However a non-linear sensor capable of 15 stops of DR compressed into 14-bit data WOULD help.

I would obviously prefer to have a linear sensor with read noise low enough to support 15 stops of DR. I think that would be much better than applying curves to the information to compress and decompress it. A non-linear sensor would just be a means of overcoming high read noise...it would just be a stop-gap measure. However Canon would have to get their read noise down to around 2e- with a 60ke- FWC (i.e. 5D IV), or 3e- with a 100ke- FWC (i.e. 1D X II), in order to have 15 true stops of linear DR. Rather doubtful they have achieved that...I might believe they have reduced read noise down to the ~10e- range @ ISO 100...that would be almost 1/4 what the 1D X had, and if the II still has around 90ke- FWC, then that would be ~13.2 stops DR (I'd take that, though!!). Sony has barely achieved that even...most FF exmors have 4-6e- read noise at ISO 100, and the A7s has ~25e- (although it also has 150ke- FWC).

RE pre-de-emphasis.. really what is the point?.. what is easier to encode, 14 bits non-linear or 16 bits linear, CR2 files already do lossless compression of 14 bits into about 9 on average, card costs are now almost zero compared to camera costs. If it were my decision I'd go with CR3 which would merely be 16bit data with TIFF compression and keep everything linear. Perhaps Canon see CR3 as being viewed as a potential problem so want everyone on side (Adobe etc) before letting that little nugget of info out (after all Pros want things to "just work" and don't care so much about specmanship).. but that's pure speculation.

It would explain them only releasing jpegs

Non-linear amplification of the analog signal is significantly more complex, requiring a considerable investment of die space per pixel. Unless Canon has shrunk to something smaller than 180nm, I don't think they even have the space to implement such a thing. It's not about what's easy, either...it's about what's effective. If Canon's read noise hasn't changed, then encoding in 16-bit buys no one anything, because instead of 15adu noise you'll have 60adu noise...it's the same amount of noise, just encoded more finely. Your still limited to the same tonal range, which before topped out at not even 4096 levels...why store less than 4096 usable levels of information in 65535 levels worth of numeric space? It's totally wasteful, would limit their ability to process at a fast rate, would increase data size (they probably wouldn't have been able to achieve 170 continuous RAW frames), would increase storage space requirements, etc.

 

[LonelyBoy]

Is higher bit-depth files something that could be released later via firmware if they're still working on everything else around it, like DPP?

 

[3kramd5]

Is higher bit-depth files something that could be released later via firmware if they're still working on everything else around it, like DPP?

If the hardware supports it (16-bit ADCs, memory registers, etc).

 

[Lee Jay]

It depends. There is a massive amount of highlight information, more than you really need. If you compress properly, what you lose isn't necessarily as important as what you gain. It's 14 bits of data, but that data could represent more than 14 stops of information.

agreed! The highlight information is nowhere near as critical as the lows....going from 11 to 13 while skipping 12 is significant, while going from 16371 to 16379 may be 4 times the jump, but the difference in an image would be almost completely undetectable.

Exactly! Hah! Glad someone gets it.

Think "floating point numbers"

The natural logarithmic encoding of floating point numbers is ideal for storing images. Lightroom can do 16 bit floating point math for a very satisfactory 30 stops or so of DR in the image.