Gary Petersen wrote:
It works nice with this full automatic default setting of Noise Ninja too.
Thanks for that Gary. It's more or less as I thought: NN may smooth a few pixels on screen in this case, but I would rather print the original.
While it's a good idea to use the least noisy camera you can afford, at these speeds I don't see removal of grain as helping printing very much. That's why I haven't bought an NR package yet.
Don, streaks are most intrusive on flat tones, like background blur, which means they can often be removed by painting in a dust and scratches layer. Ideally one would not have to, but it's a workable last resort. Your child's face has faintly visible streaking too, just like my shot of the guitarist. That's trickier to fix. I would love to find a noise reduction package which can deal with streaks automatically, but I fear it doesn't exist.
I hasten to add that no streak correction was done on my shot of the guitarist.
simonella_viru wrote:
what do you think is a fair comparison between a super-mega-pixel camera like the 1ds3 and a camera like the d3 that has about half of the pixels?
The best way is to uprez the D3 file to 21 megapixels, but no doubt many people won't accept that doing that gives you the closest possible approximation to comparison by huge print. Downrezzing the 1Ds3 file is unfair because you are throwing away its detail.
In the days when M8 performance was contentious in the alt forum, many people seemed to think that something was lost from an M8 file when it was uprezzed, so although I don't subscribe to that piece of mysticism, to satisfy the sharp pixel crowd, I uprezzed the M8 file to 40 megapixel using nearest neighbour interpolation, so that the file was identical to the original in every way, and uprezzed the 5D file to the same size using bicubic.
In the case of comparing D3 and 1Ds3 one would need to double the base of the D3 file using nearest neighbour, giving 48 megapixels, but a visually identical file, and then match the new baselength using bicubic in the 1Ds3 file, to also give a 48 megapixel interpolated file.
Personally I don't think the second method shows you anything that the first method doesn't.
Thanks Richard, I'm pleased to have you declare my camera normal. I was worrying a bit (especially after an earlier banding scare caused by a duff CF card).
I know I've gone on about Bibble with you before, but hear me out on this one.
Sean Puckett makes a plugin for Bibble that I think you would be very interested in for the purposes of pushing. He has developed an exposure amplification algorithm that can recover detail with dramatically less noise amplification that typical tools.
Glassbottle wrote:
Thanks Richard, I'm pleased to have you declare my camera normal. I was worrying a bit (especially after an earlier banding scare caused by a duff CF card).
Nice trick with the Dust&Scratches there.
I am in the same boat. Initially I was totally amazed at the posted 12800 shots, because my 1Ds3 doesn't seem to be able to get anything like that. But with explained technique I feel better now
Let me make this clear - I did not perform selective dust and scratches on the guitar player. That image is what I get straight out of DPP, and has only had a very gentle dust and scratches over the whole frame, because I'm too cheap to buy NR software. That level of d+s would be invisible in most normal print sizes, so to all intents and purposes, that's what I get straight outta DPP.
My thanks to brainiac who gave me courage to buy my 1Ds Mk III after he exposed my faulty reasoning regarding relative noise between the Mk III and 1Ds MKIII. * As for D3 verses the Mk III s. I rate the D3 as up to 1 stop more "effective" at hi ISO (D3=6400 and Mk IIIs =3200). YMMV!
* I thought the Mk III noise should be better than the 1ds MKIII but just as brainiac sez they are identical by my informal test: When both cameras capture the same image framing (with a 28mm on the Mk III and a 35mm on the 1Ds Mk III). After enlargement to 500% the noise is virtually identical.
SoundHound wrote:
I rate the D3 as up to 1 stop more "effective" at hi ISO (D3=6400 and Mk IIIs =3200).
But how are you doing that comparison? To make the comparison fairly, you need to...
a) take the same picture under the same conditions at the same shutter speed and aperture
b) making sure that you use the same lens, as f2 on one lens doesn't necessarily transmit the same amount of light as f2 on another
c) process out of the latest DPP with no sharpening or noise reduction
d) apply Neat Image or some other noise reduction filter to the Canon file, and if necessary, the D3 file
e) uprez the D3 file to 21 megapixels and compare on screen
You might also test by printing the two images to the same size.
If you didn't do it that way, there's little point in zooming in on pixels and trying to assess which is noisier, as that won't tell you which camera is more usable at iso 6400, it will only tell you how noisy a pixel is.
Gary even so, the default NN profiles are less effective if you use LR/ACR because of the amount of noise the these engines produce. If you want the best high ISO performance, use DPP.
If you want to use ACR/LR then you need to profile each image individually. Or make your own set of profiles through LR/ACR.
Sorry, its a bummer I know but those is the facts :-(
I just printed two 8x10s of my ISO 3200 shots. That's as large as I can print at home. One from DPP with no noise reduction or sharpening and the other from photoshop after using Noise Ninja. I see no noise at all in either one. I declare the 1D Mark III perfectly usable at 3200.
Might it make sense for Canon to provide the 1DsMkIII with a firmware update that allows ISO 6400 and 12800 (assuming it could be done with FW)? They might fear that pixel peepers would consider the results too noisy, but the sheer number of pixels obviously allows the camera to produce very usable prints at those exposures, so why leave it necessary to use exposure compensation to get the exposure index we want?
I use a 1DIII, not a 1DsIII, but from looking at other's results, it seems to me that ISO 12800 shots from a 1DsIII would make better prints of any given size than ISO 3200 shots from the original 1D, and maybe better than ISO 1250 shots from the original 1Ds.
I was following this discussion with great interest. The main thing that I don't get is people using DPP and getting great results. I find that DPP gives the most atrocious results, especially when it comes to noise. Then I see that other people have better luck. So let me ask: this is an image from last night, 1Ds3, ISO 3200, not what I would call underexposed / pushed. First one with DPP (newest version - no NR and sharpening, as stated in the "directions" in this thread), then in ACR, with no luma NR and 25 chroma NR (my standard settings). Maybe it's just me, but I find the DPP results ... well, atrocious. The ACR image has been white balanced and has a custom camera curve applied, but that should not explain the shocking difference.
Stanj, turn all noise off in LR/ACR. Match the WB, it does have an effect on visible noise, I can see your wb is off a bit. You are also comparing a NR reduced image to a non NR reduced image.
At any rate, you still need to remove the noise in Noise Ninja afterwards. I didnt say that the noise wasnt there, it is, its just that it has less of an impact and IQ.
