ChrisDM wrote:
...if you need to shoot at 12000 ISO, until the 5DMk2 comes out your only choice is one of the Nikons.
That's the only bit I don't agree with. If you are happy shooting raw, and applying appropriate NR, then you won't see much difference between the Nikons and the 5D2 or 1Dmk3's at isos 6400 and 12800 in low light.
RalphJ wrote:
Brainiac, that's very impressive. Are there any special PP techniques we should emphasize to achieve that kind of result, or is it just the usual combination of RAW tools?
I convert from the latest DPP using no sharpening and no NR. Then you can apply NeatImage or one of the other NR tools to taste. Recently I have been experimenting with a noise reduction method that works very well, as follows:
- duplicate base layer
- run smart blur on the dupe layer
- set smart blur settings to 8, 16, 24 or 32 pixels, high quality, overlay edge mode, and set threshold so that the right quantity of edges show in the preview (the white lines will be where you are not removing noise/colour detail)
- use wand, tolerance 0, non-contiguous, not all layers to select all white pixels
- delete
- feather by a pixel or so
- delete again two or three times
- gaussian blur by 2 pixels
- set layer blending mode to colour
- set layer blending options to blend if grey 0, 100/200 or thereabouts (option click the white point slider to split it into the upper and lower graduation points), so that only the shadows are affected
- if necessary, un-apply by painting black on a layer mask
- if necessary, apply a 1 pixel radius dust and scratches with threshold between 20 and 50 to smooth out grit, but be careful as this may be throwing away detail too
Obviously this can all be made into an action so it's one click.
It can also help to apply dust and scratches separately to each channel with different parameters, as, depending on the light source, each channel usually ends up with a different scale and severity of noise.
If you apply sharpening at all, apply it to a duplicate layer in luminosity blending mode. That means that only shade is sharpened, and colour differences aren't.
Sometimes, for portraits and such, I make a noise reduction layer a little too strong, and then give it a layer mask and carefully fade it where I want more bite, like in the eyes.
That's the only bit I don't agree with. If you are happy shooting raw, and applying appropriate NR, then you won't see much difference between the Nikons and the 5D2 or 1Dmk3's at isos 6400 and 12800 in low light.
Actually I think we agree. I don't see much difference in the 1DMk3 or 5D2 samples versus the Nikons either, except for the fact that some cameras are designed to shoot these ultra high ISOs and some are not. However there's no better method for controlling noise than acquiring a properly exposed capture. Pushing underexposed images is not the most effective practice, especially when the ISOs are high.
I really like the bird shot but is it just me or are all the 5D2 samples really soft at the per pixel level? I haven't seen a shot that doesn't look like its been hit with NR and detail has been smeared away. They seem to lack a certain "crispness". Not just the bird shot but also Vincent Laforet's shots, DPReviews shots, everything. They all look very soft at 100%. 21 megapixels is not the reason given the pixel density is the same as a 8MP APS-C sensor...
philber wrote:
You are one person, Richard, maybe not quite enough to be a majority
So maybe the rest of us can focus on conditions which more closely replicate the every day use of our cameras?
It's not just brainiac that uses high isos, most indoor event photographers would tend to push their cameras in this way too. High ISO also represent the every day use of my camera too. My cameras are set at ISO 1600 as a default and I often push up to 3200 in PP. Actually I can't remember when I've ever used ISO 100 except when using the 85L outdoors
Dawei Ye wrote:
I really like the bird shot but is it just me or are all the 5D2 samples really soft at the per pixel level? I haven't seen a shot that doesn't look like its been hit with NR and detail has been smeared away. They seem to lack a certain "crispness". Not just the bird shot but also Vincent Laforet's shots, DPReviews shots, everything. They all look very soft at 100%. 21 megapixels is not the reason given the pixel density is the same as a 8MP APS-C sensor...
It's not just brainiac that uses high isos, most indoor event photographers would tend to push their cameras in this way too. High ISO also represent the every day use of my camera too. My cameras are set at ISO 1600 as a default and I often push up to 3200 in PP. Actually I can't remember when I've ever used ISO 100 except when using the 85L outdoors...Show more →
The 5dmk2 jpegs are NRed to death is really horrible how much it destroys detail. I took some raw samples from 5dmk2 and processed them with dpp 3.5.1 and applied NR with noise ninja in CS3 and the diff is huge. But the original files are extremely noisy at iso3200+ and there is noise even at iso 200. If it were not for the 21mpx I would say they did not improve much on the 5dmk1 noise.
ChrisDM wrote:
Pushing underexposed images is not the most effective practice, especially when the ISOs are high.
It's exactly the same thing as pushing ISO in the camera, as long as you have gone past the analogue amplifying maximum. In modern cameras, it is not even sure that there are any levels of analogue amplifying, just base-ISO and then digital cranking. Just as you do in your RAW developer.
I don't know about Canon, but there is very little difference if you underexpose one stop and the crank it equally afterwards with the D700, even at ISO 6400.
Alot of people don't understand that you can shoot ISO 25600 (or whatever) with whatever camera that produces RAW files. Just underexpose and crank it back in PP, and you've got your insane ISO.
What you don't get is the wanted (?) in-camera higher noise reduction at higher ISO speeds, often "needed" because the pictures will look like crap without it.
Edit: Sorry, I just realized that this has been discussed earlier in the thread.
Dawei Ye wrote:
I really like the bird shot but is it just me or are all the 5D2 samples really soft at the per pixel level? I haven't seen a shot that doesn't look like its been hit with NR and detail has been smeared away. They seem to lack a certain "crispness". Not just the bird shot but also Vincent Laforet's shots, DPReviews shots, everything. They all look very soft at 100%. 21 megapixels is not the reason given the pixel density is the same as a 8MP APS-C sensor...
Out of camera jpegs have a lot of NR, but RAW captures are as detailed as (if not better than) my 40D and 20D shots at pixel level.
Makten wrote:
It's exactly the same thing as pushing ISO in the camera, as long as you have gone past the analogue amplifying maximum. In modern cameras, it is not even sure that there are any levels of analogue amplifying, just base-ISO and then digital cranking. Just as you do in your RAW developer.
I don't know about Canon, but there is very little difference if you underexpose one stop and the crank it equally afterwards with the D700, even at ISO 6400.
Alot of people don't understand that you can shoot ISO 25600 (or whatever) with whatever camera that produces RAW files. Just underexpose and crank it back in PP, and you've got your insane ISO.
What you don't get is the wanted (?) in-camera higher noise reduction at higher ISO speeds, often "needed" because the pictures will look like crap without it.
....Show more →
It's not "exactly the same", for the primary reason that you can't see what you're getting! When shooting two stops under and then pushing in post you don't know what you've got 'til you get home. Sounds like the good old days... It makes a lot more sense to use a camera that shoots at 6400 to 12800 if in fact that's where you shoot. In fact the ability to go to 6400 is one of the primary reasons I shoot weddings/concerts etc with my 1D3 instead of my 1Ds3.
brainiac wrote:
Yes, I would, but ONLY if you remember brainiac's rule: always uprez files to the same number of megapixels before comparing 100% crops.
If you forget to observe brainiac's rule, then comparing 100% crops is positively the worst way of comparing the sensor noise from two cameras, because it is reliably and methodically biased in favour of the lower resolving camera.
on "Brainiac's Rule" (which I think is a pretty good one BTW).
alek
brett maxwell wrote:
What are the thoughts on the 5D2 using sRAW for 10mp? 10mp is plenty for 98% of what I shoot, but will the pixel advantage for ISO still apply?
Apparently not. The initial impression of sRaw is that it stil does not take advantage of down resing in a way that can reduce noise. But it will still be a nice file. Above 1600 or 3200 I would plan to shoot regular raw.
RalphJ wrote:
Brainiac, that's very impressive. Are there any special PP techniques we should emphasize to achieve that kind of result, or is it just the usual combination of RAW tools?
Sorry, I find luminosity noise unnecessarily sharpened for a portrait. Although as a print it would be softened somewhat.
ChrisDM wrote:
It's not "exactly the same", for the primary reason that you can't see what you're getting! When shooting two stops under and then pushing in post you don't know what you've got 'til you get home.
Two stops isn't going to make your pictures so dark that you can't see if they are properly focused or stuff like that.
I do this kind of often with the D700; "overriding" my normal auto-ISO maximum (6400) by simply turning the exposure compensating if the light still isn't enough for proper shutter speeds at ISO 6400.
dcmiller wrote:
Sorry, I find luminosity noise unnecessarily sharpened for a portrait. Although as a print it would be softened somewhat.
No need to apologise. This would print quite well, in my experience. It hasn't been sharpened much, if at all, but it was taken with a Zeiss 28 at f5.6, which I find makes the 450D go ping.
musclepics wrote:
The originally posted 5D2 shots look good, but I actually get cleaner ISO3200 shots with my 1D3
I won't be surprised that the 1D3 with its bigger pixels will have lower noise at ISO 3200 than the 5D2, when compared pixel-to-pixel.
The 5D2 has a similar sensor to the 1Ds3, and DxO's tests show that at the pixel level the 1D3 is better, while the 1Ds3 takes the edge on a frame-to-frame comparison.
on "Brainiac's Rule" (which I think is a pretty good one BTW).