cgarcia
Offline
• •
Upload & Sell: Off
|
 p.1 #1 · Canon 1DX2 dynamic range analyzed from RAWs | |
The 1DX2 at ISO 100 improves by 1.5 EV the 1DX dynamic range. It beats the 5DS by about 1 EV. It also beats by about 0.33 EV the 80D. However, at high ISO it merely manages to catch the old 1DX, and at the highest ISO levels it falls behind by about 0.3-0.5 EV, due to higher read noise.
Interestingly, the 1DX2 still shows a noticeable DR drop at low ISO (much more marked than the 80D) but not so bad compared to the old sensor tech. Between ISO 800 and ISO 100, the 80D dynamic range increases 2.13 EV, while the 1DX2 only goes up by 0.97 EV.
https://lightful.github.io/talk/photo/sensor/1DX2_1DX_80D_7D2__DR.png
In summary, those shooting the 1DX mainly at ISO 3200 or above, shouldn't immediately sell it: at high ISO, the 1DX2 has a bit more noise (more read noise, exactly the same photon noise -physics laws- and likely the same thermal noise due to the short exposures). But shooting RAW at low ISO the story is very different, and there is a very noticeable improvement in the shadows recovery ability. Nevertheless, high ISO noise "quality" may also have improved (was already good) and JPEG quality could be better thanks to the new processor, so your mileage may vary (not to mention the other camera features besides the sensor). After looking at the 80D data, I must confess that personally was expecting better results in the 1DX2, but that doesn't means that these are not great.
Both the data bellow and the chart above are fully consistent and comparable (all the represented cameras are analyzed exactly the same way); in fact, the 7D2 chart changes slightly compared to the one published in the 80D analysis because I have calculated all the plotted ISO levels instead of interpolating the missing ones. Note in the 1DX2 chart the singularity at ISO 400. It seems a real characteristic, because two different shots (taken with 6 hours of difference, one properly exposed and the other severely underexposed) corroborate exactly the same result. As with ISO 400, for some other ISO levels (3200, 25600, 51200) I had several pictures, and all were analyzed to get more confidence in the results.
Note: all the RAWs (except for the 1DX2) come from Imaging Resource ISO series (the 1DX2 have been received from another source which I won't disclose; please do not ask me for these RAWs).
ISO 100
1DX2 - DR 13.55 at 8MP (12.89 at 20MP), read noise 2.0845 DN
1DX - DR 12.01 at 8MP (11.42 at 18MP), read noise 5.19861 DN
ISO 200
1DX2 - DR 13.40 at 8MP (12.74 at 20MP), read noise 2.31997 DN
1DX - DR 12.07 at 8MP (11.49 at 18MP), read noise 4.98757 DN
ISO 400 (2 samples)
1DX2 - DR 12.81745 at 8MP (12.15645 at 20MP), read noise 3.14005 DN
1DX2 - DR 12.83568 at 8MP (12.17468 at 20MP), read noise 3.10066 DN
1DX - DR 12.02080 at 8MP (11.43580 at 18MP), read noise 5.17441 DN
ISO 800
1DX2 - DR 12.58 at 8MP (11.92 at 20MP), read noise 3.68397 DN
1DX - DR 11.93 at 8MP (11.34 at 18MP), read noise 5.50986 DN
ISO 1600
1DX2 - DR 11.82 at 8MP (11.16 at 20MP), read noise 6.27188 DN
1DX - DR 11.71 at 8MP (11.12 at 18MP), read noise 6.41935 DN
ISO 3200 (2 samples)
1DX2 - DR 11.2831 at 8MP (10.6221 at 20MP), read noise 9.09559 DN
1DX2 - DR 11.2646 at 8MP (10.6037 at 20MP), read noise 9.21231 DN
1DX - DR 11.2559 at 8MP (10.6709 at 18MP), read noise 8.79265 DN
ISO 4000
1DX2 - DR 10.94 at 8MP (10.28 at 20MP), read noise 11.5206 DN
ISO 6400
1DX2 - DR 10.44 at 8MP (9.78 at 20MP), read noise 16.2208 DN
1DX - DR 10.50 at 8MP (9.92 at 18MP), read noise 14.8059 DN
ISO 12800
1DX2 - DR 9.48 at 8MP (8.82 at 20MP), read noise 31.7281 DN
1DX - DR 9.77 at 8MP (9.18 at 18MP), read noise 24.6235 DN
ISO 25600 (3 samples)
1DX2 - DR 8.49289 at 8MP (7.83192 at 20MP), read noise 62.9152 DN
1DX2 - DR 8.49692 at 8MP (7.83596 at 20MP), read noise 62.7387 DN
1DX2 - DR 8.49736 at 8MP (7.83639 at 20MP), read noise 62.7207 DN
1DX - DR 8.79292 at 8MP (8.20796 at 18MP), read noise 48.4779 DN
ISO 51200 (3 samples)
1DX2 - DR 7.52679 at 8MP (6.86583 at 20MP), read noise 122.895 DN
1DX2 - DR 7.53064 at 8MP (6.86967 at 20MP), read noise 122.570 DN
1DX2 - DR 7.52280 at 8MP (6.86184 at 20MP), read noise 123.234 DN
1DX - DR 8.05955 at 8MP (7.47459 at 18MP), read noise 80.6098 DN
This data is very similar to the DXO estimation, but not totally comparable. For example, DXO reports 8.08 EV at ISO 25600 for the 1DX, while the value here is 8.79 EV (this is a extreme case, note that usually the offset is a lot smaller). Here we use the DXO definition of DR, but calculate it directly against the noise floor "fingerprint" recorded by the own sensor (instead of plotting several SNR curves to infere it). So I have more confidence on these results (however, DXO calculates a lot more sensor performance metrics).
The 1DX2 black levels are identical to the 80D (about 512 at ISO 100-200, and 2048 for the rest of ISO levels). Interestingly, had the ISO 400 used a 512 black level, the DR singularity in the chart would almost have dissapeared (but indeed does is 2048). I have assumed a saturation of 16383 DN for a few ISO levels not overexposed (1DX2 ISO 100 DR could be about 0.1 EV less in the worse case, but 1DX reached such white level anyway). The 1DX2 masked pixels area is not huge (72 pixels at the left and 38 at the top, exactly as it is in the 7D2) which partially breaks the recent trend of increasing it (the 1DX used 126 at the left and 100 at the top).
PS: I'm in the final stage of cleaning up the source code of the custom tool I use for the analysis (currently still requires manual editing and a compiler) and in a few days I will upload and document it on Github (also with a compiled executable) in order to allow anybody using or improving it. It has more useful features to characterize the sensor performance (which require to take special shots, though).
Edited on Apr 03, 2016 at 05:21 PM · View previous versions
|
