Simultaneous dual gain sensors

[Patrick B.]

30 minutes ago, thebrothersthre3 said:

Couldn't RAW potentially yield higher dynamic range or more highlight recovery. 

yeah I wondered this too..  but usually in their tests the compressed codecs win out for DR because they already have pretty optimized NR happening.  and I’m not sure if Canon’s CRL has the same kind of highlight recovery feature as BRAW.. but would have been interesting to at least see the raw results to compare.  I guess he said they were rushed so maybe that’s why they didn’t.

[androidlad]

59 minutes ago, Patrick B. said:

yeah I wondered this too..  but usually in their tests the compressed codecs win out for DR because they already have pretty optimized NR happening.  and I’m not sure if Canon’s CRL has the same kind of highlight recovery feature as BRAW.. but would have been interesting to at least see the raw results to compare.  I guess he said they were rushed so maybe that’s why they didn’t.

Since the Alexa was tested using ProRes codec for 14 stops of DR, it's only fair to test compressed codec for all other cameras as a standardised metric.

[thebrothersthre3]

1 minute ago, androidlad said:

Since the Alexa was tested using ProRes codec for 14 stops of DR, it's only fair to test compressed codec for all other cameras as a standardised metric.

They always test blackmagic with RAW though so I don't see why not with anyone else.

[androidlad]

1 minute ago, thebrothersthre3 said:

They always test blackmagic with RAW though so I don't see why not with anyone else.

BRAW is not RAW, it's a compressed, partially debayered codec.

[thebrothersthre3]

2 minutes ago, androidlad said:

BRAW is not RAW, it's a compressed, partially debayered codec.

They've done tests with Cinema DNG as well. They rated the Ursa Mini at above 13 stops using Cinema DNG (NR and Highlight recovery). 

[Patrick B.]

1 hour ago, androidlad said:

Since the Alexa was tested using ProRes codec for 14 stops of DR, it's only fair to test compressed codec for all other cameras as a standardised metric.

yeah good point.  it’s interesting that Alexa was the only camera that matches their stated DR as measured by Cinema5D.  then again I guess other companies are using a low S/N ratio likely in their testing to get higher number whereas as ARRI are more conservative and realistic.

curious how Red would stack up using their standard methodology as I know they have very good DR.

[thebrothersthre3]

1 hour ago, Patrick B. said:

yeah good point.  it’s interesting that Alexa was the only camera that matches their stated DR as measured by Cinema5D.  then again I guess other companies are using a low S/N ratio likely in their testing to get higher number whereas as ARRI are more conservative and realistic.

curious how Red would stack up using their standard methodology as I know they have very good DR.

RED is probably the worst when it comes to dynamic range claims. They say things like 21 stops of dynamic range lol. The latest RED offerings come pretty close to ARRI though, the Monstro has very clean shadows. 

[ntblowz]

The result from newshooter said otherwise, but they shoot in raw not compressed like C5D do.

C500 ii is very noisy in the shadow area vs c300 iii with DSG.

 

[thebrothersthre3]

Yeah need to see some good dynamic range tests in RAW

[KnightsFan]

13 hours ago, Patrick B. said:

Yeah, just read that!  I’d be lying if I said I wasn’t disappointed.  Not that I’m gonna pick up a C300 MK III.. but I’d just like to think after 10 years someone else can match an ARRI for DR ha ha.  I guess at this point everybody is pretty close though.

So I guess ARRI’s implementation in the ALEV III is very special.. not to say they are the same approach since they don’t have the split, dual-pixel design.  But I thought surely that using two separate exposures would yield even greater DR.  Puzzling...

The 14 stops that C5D measured on the Alexa was at 2k resolution, and presumably the C300 III number of 12.8 would be closer to ~13.5 if downscaled to 2k. So it seems very close in terms of SNR.

[cardinalallen]

On 4/21/2020 at 2:52 PM, androidlad said:

C300 MK III DGO only scored 12.8 stops using ARRI standard.

https://***URL not allowed***/canon-c300-mark-iii-lab-test-pre-production-dynamic-range-rolling-shutter-and-latitude/

For reference:

C300 MK II: 12.3 stops

C500 MK II: 13.1 stops

I think there are a few caveats with this test.

First, it's using a pre-production camera.

Second, it's using XF-AVC. The author argues that with a log curve like CLog2, there won't be any difference between that and raw. Personally, I suspect it's a bit more complicated than that, and depends significantly on the implementation. I've heard for example that the Alexa ekes out a tiny bit more DR with arriraw than with Prores. 

Third, and related to the first and second point, Clog2 is only rated for 15-stops DR. Whilst under the above measurement the C300 Mk III only has 12.8 stops, it technically is rated by Canon at 16-stops. This matters because presumably some work has to happen on Canon's side to figure out how to distribute its DR within the Clog2 gamma. 

In the comments, the author writes the following: 

Quote

Interestingly, IMATEST also calculates the absolute number of patches that it can detect – and here the C500 MKII shows 15.8 vs the 16.8 that are calculated for the C300 MKIII. So in a way the C300 MKIII sees “more” information buried in the noise.

Given that it's a pre-production camera, there could well be some tweaking that needs to be done on this front.

Fourth, it says nothing of the actual roll-off. This is ultimately the crucial question. Canon said that the DGO would allow for higher saturation in highlights, which is crucial to that smooth roll-off; what happens at present is that different colour channels clip at different rates and thus you end up with colour posterisation in the highlights. Roll-off always looks so much better in black-and-white for this reason.

This test might prove to be accurate with the production model, but I think we should treat it with caution.

[Patrick B.]

@cardinalallen good information to consider.  I overlooked the comment about the C300 seeing more into the shadows.  So even though the SNR1 rating was lower than the C500, it may be that you can actually eke out more in post.  Interesting point about the color channels in highlights.  I’ll be interested in seeing how it handles this.

Also, I noticed the Mavic Air 2 does some kind of HDR processing where different pixels are mapped to different exposures, similar to what DGO is doing.  Are other cameras doing this with Sony sensors?  That sounds like a useful feature for high resolution sensors that I didn’t consider before.  I’ve heard talk about quad Bayer sensors where blocks of pixels can become pseudo bigger pixels and record better low light at lower resolutions like we see in new camera phones.  But I never considered that high resolution sensors could actually record two exposures in lower resolution and essentially turn what is usually a negative sensor attribute for video into a positive.

[androidlad]

9 hours ago, cardinalallen said:

I think there are a few caveats with this test.

First, it's using a pre-production camera.

Second, it's using XF-AVC. The author argues that with a log curve like CLog2, there won't be any difference between that and raw. Personally, I suspect it's a bit more complicated than that, and depends significantly on the implementation. I've heard for example that the Alexa ekes out a tiny bit more DR with arriraw than with Prores. 

Third, and related to the first and second point, Clog2 is only rated for 15-stops DR. Whilst under the above measurement the C300 Mk III only has 12.8 stops, it technically is rated by Canon at 16-stops. This matters because presumably some work has to happen on Canon's side to figure out how to distribute its DR within the Clog2 gamma. 

In the comments, the author writes the following: 

Given that it's a pre-production camera, there could well be some tweaking that needs to be done on this front.

Fourth, it says nothing of the actual roll-off. This is ultimately the crucial question. Canon said that the DGO would allow for higher saturation in highlights, which is crucial to that smooth roll-off; what happens at present is that different colour channels clip at different rates and thus you end up with colour posterisation in the highlights. Roll-off always looks so much better in black-and-white for this reason.

This test might prove to be accurate with the production model, but I think we should treat it with caution.

The only caveat I agree with is the fact that it's a pre-production camera.

Remember it was a standardised lab test, Alexa was tested in ProRes using the same methodology, so it's completely fair to test C300 MK III with XF-AVC. With Alexa being the benchmark, there's something to compare to, so if want to take into account DR distribution in specific log curves, highlight roll-off etc., then the Alexa DR would change too, so maybe instead of 14 vs. 12.8, you could get 15 vs. 13.8. Again remember it's a standardised test, the DR numbers are all relative to the benchmark, we can't look at them on their own.

[androidlad]

3 hours ago, Patrick B. said:

@cardinalallen good information to consider.  I overlooked the comment about the C300 seeing more into the shadows.  So even though the SNR1 rating was lower than the C500, it may be that you can actually eke out more in post.  Interesting point about the color channels in highlights.  I’ll be interested in seeing how it handles this.

Also, I noticed the Mavic Air 2 does some kind of HDR processing where different pixels are mapped to different exposures, similar to what DGO is doing.  Are other cameras doing this with Sony sensors?  That sounds like a useful feature for high resolution sensors that I didn’t consider before.  I’ve heard talk about quad Bayer sensors where blocks of pixels can become pseudo bigger pixels and record better low light at lower resolutions like we see in new camera phones.  But I never considered that high resolution sensors could actually record two exposures in lower resolution and essentially turn what is usually a negative sensor attribute for video into a positive.

The HDR video mode on Mavic Air 2 uses the DOL-HDR feature on the newer Sony sensors, it's very similar to dual gain but the two readout are not completely simultaneous (a few miliseconds apart), and requires 2 ADC for high and low gain, which halves the max frame rate and doubles the rolling shutter.

[Patrick B.]

58 minutes ago, androidlad said:

The HDR video mode on Mavic Air 2 uses the DOL-HDR feature on the newer Sony sensors, it's very similar to dual gain but the two readout are not completely simultaneous (a few miliseconds apart), and requires 2 ADC for high and low gain, which halves the max frame rate and doubles the rolling shutter.

Interesting.  So there is still a penalty for this increased range.  I wonder if the ALEV III uses 2 ADC.  The description makes it seem like both readouts are fed through one ADC, but not sure.  I guess frame rate constraints and increased readout times could be more desirable than the motion changes that result from the alternating frames approach in HDR/WDR modes of cameras like Red and Z Cam.