New ISO-comparisons.

[andybarton]

Advertisement (gone after registration)

I am not saying it invalidates any test, but you have a lot of variables in here.

 

You are obviously using two different bodies.

 

You are using two different lenses, designed for different types of camera, from different manufacturers. One at the very top end (albeit, using their entry level model), one from the lower end of the SLR lens market.

 

You have also upres'd the Nikon image, which is bound to have an effect. Why not just post 100% crops, rather than try to make the Nikon image the same size as the Leica one?

 

I do understand what you are trying to prove here, and as a D700 owner, I do know what they are like, but it would be a more valid test, IMHO, if you used two Leica lenses of similar age and design. AFAIK, a 50 Summicron on BOTH cameras would be a better comparator.

 

The fewer variables one introduces, the more valid the test. Given that you are testing ISO from the chip, you should try to make everything else equal.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[patrick parker]

There is no other other digital rangefinder camera on the market that comes close the M9,

it's the most quite camera in it's league, but it's two steps behind on the noise factor which is a real shame. Is it really such a big deal to put sensor in there that is comparable to what canon or nikon used couple of years ago? Do you think if they hadn't suffered financially M9 would have a better sensor now?

[diogenis]

There is no other other digital rangefinder camera on the market that comes close the M9,

it's the most quite camera in it's league, but it's two steps behind on the noise factor which is a real shame. Is it really such a big deal to put sensor in there that is comparable to what canon or nikon used couple of years ago? Do you think if they hadn't suffered financially M9 would have a better sensor now?

 

It is also two steps above in IQ.

What do you prefer? lose IQ in favor of a questionable denoise routine IN the camera, or get a perfect file and use a super duper denoise software on it and get even better files in the end?

The Japanese know how to sell sure... but when you compare A-B pics they lose...

[diogenis]

I tried quickly with dfine and noiseware on the 2500 ISO imag. This is better indeed but their is less details than in the Nikon files then. And even some good sharpening cannot get them back.

 

But of course, this is Sunday morning where I am so it is probably possible to do much better than I did.

If someone can and is kind enough to post its results, that would be good.

 

I believe it would be best to try that on the original RAW and UNCOMPRESSED files so that you bypass whatever filters, denoise, de-anything JPEGs introduce. After all this is what CaNikon do with their cameras.

 

 

Edit: It's quite obvious why Leica chose the route of leaving the picture unharmed. With the way s/w evolves you can get amazingly powerful tools easily, instead of using whatever stationary hardware circuit they use to filter noise on cmos sensors

[ulrikft]

I am not saying it invalidates any test, but you have a lot of variables in here.

 

You are obviously using two different bodies.

 

You are using two different lenses, designed for different types of camera, from different manufacturers. One at the very top end (albeit, using their entry level model), one from the lower end of the SLR lens market.

 

You have also upres'd the Nikon image, which is bound to have an effect. Why not just post 100% crops, rather than try to make the Nikon image the same size as the Leica one?

 

I do understand what you are trying to prove here, and as a D700 owner, I do know what they are like, but it would be a more valid test, IMHO, if you used two Leica lenses of similar age and design. AFAIK, a 50 Summicron on BOTH cameras would be a better comparator.

 

The fewer variables one introduces, the more valid the test. Given that you are testing ISO from the chip, you should try to make everything else equal.

 

I'll try to adress your points one at a time.

 

1) The Sigma 50 1.4 is not a low end lens, it is among the sharpest 50 1.4s out there and is more similar to the canon 50 1.2 than the canon/nikon 50 1.4s. I could try to do a d3x + sigma 50 1.4 vs. M9 + summarit 50 2.5 to convince you, but that would have to be at a later date. My firm opinion is that the sigma 50 1.4 is as sharp as the summarit 50 2.5 at f/8.

 

2) I have explained (and linked to an article about) why I upres. Comparing isos at 100% without resizing down or up, will give the higher megapixel sensor a disadvantage. I argue that the disadvantages that upressing introduces, are far less than what comparing at 100% pixel view without upressing (or downressing) introduces.

 

3) I do have a leitax-adapter, so I could have used a 50 2.0 Leica R-lens, but I honestly believe that the Sigma is a better performer in many ways. I also have nikon, minolta and other fast 50s for my camera, but so far the only one coming close to the sigma 50 1.4 is the minolta rokkor 58 1.2, which is unsuited for many different reasons for this test (wrong focal lenght for one)

 

I think I'm doing the tests and procesing the data in the best possible way with the gear I have at hand. If someone want to send me a D3x and different lenses, I would love to review that too, but I don't have that gear myself, so that would be impossible to do right now.

[jrgeoffrion]

I've done a few iso comparions more, in addition to day-to-day use testing.

I'll just link to the full files themselves (nikon d700 upressed to 18mpix). I would say that the M9 is between 0.5 and 1 stop behind the D700 noisewise, but that it keeps the detail very well in spite of the noise. No noise reduction exept the default lightroom done to any of the files. I have upressed the D700 to the M9-size of files, beacause i think that the margin of error when comparing two different filesizes on a pixel level, is larger than the errors that resizing can induce. I understand that some people disagree on this point, but I prefer this method.

 

Thanks for the comparison.

 

A few random points...

 

1. The comparison should indeed be done on a pixel-by-pixel basis to eliminate the enlargment factor when making a print (hence why I asked how Leica defined "one stop better noise performance").

2. Uprising the D700 from 4,256 to 5,212 pixel will introduce very minor image degradation and should be done on a 16-bit TIFF for optimum results and then saved as a JPG.

3. The best processing stage to remove noise is at the RAW conversion. Hence, for best results, RAW conversion settings should be optimized to remove as much noise as possible while retaining details rather than keeping the default settings and trying to remove noise in a second processing stage with software such as Noise Ninja or Neat Image.

4. Noise can actually give the impression of sharpness. I used this often when printing. It also remove most posterization due to uprising in the 8-bit domain on smooth image surfaces. Try adding noise to the D700 image and it will a lot sharper.

5. Since 50mm Sigma used in the test is one sharp lens (I have one) with optimal corner-to-corner sharpness at f/8. Here is an sharpness graph: dpreview.com - Lens Review - Fullscreen

6. In camera noise reduction settings may affect D700 noise and sharpness levels.

 

A few observations...

1. At 400ISO, I prefer the Leica.

2. At 2500ISO, I prefer the D700.

 

I'd love to have access to the RAW files to see what information each file contains...

 

I'd love to take my 5DII with the 35/1.4L, 50/1.2L, and 85/1.2L and compare it to the M9 with the 35/1.4 ASPH, 50 NOCT, and 75/1.4 and 90 APO...

[lct]

Advertisement (gone after registration)

...I could have used a 50 2.0 Leica R-lens, but I honestly believe that the Sigma is a better performer in many ways...

Then you're comparing lenses not cameras. If exact same lenses are not available, Andy's idea (Summicrons 50) is the only valid option IMHO.

[jrgeoffrion]

Then you're comparing lenses not cameras. If exact same lenses are not available, Andy's idea (Summicrons 50) is the only valid option IMHO.

 

According to the link I posted above, the Sigma outresolves the D700 sensor from the center until about 95% of the corners at f8 (at least until the very, very corners). Hence, the lens is not a factor in the D700 image...

[jackal]

jesus... now we arguing about whether lens selection invalidates the test or not. This place is so full of pedantic bores. Do some of you guys actually take pictures or do you just like to revel in your own bombastic self-aggrandising empircal reasoning ?

 

I did some rough tests up there (as did others) that pretty much prove that anyone whinging about M9 noise is some sort of obsessional gearhead who is so focussed on pixels and tech and cleanliness and sharpness, that they've long since forgotten what photography is all about.

 

Every top line camera on the market today is a great product and more than capable of serving the hobbyist and the professional; there is more choice than ever before. With something like the M9, bottom line is YOU will be the weakest link the chain long before the camera so get out there and keep shooting if you genuinely want to move the game on.

[lct]

There are well known rules to do comparisons. Avoiding variables is one of them. Is this LUF or dpreview here.

[ulrikft]

There are well known rules to do comparisons. Avoiding variables is one of them. Is this LUF or dpreview here.

 

As someone said above, the Sigma outresolves most of the D700 sensor at F/8, it will not introduce any variables that are very relevant to the test.

[egibaud]

I am not saying it invalidates any test, but you have a lot of variables in here.

 

You are obviously using two different bodies.

 

You are using two different lenses, designed for different types of camera, from different manufacturers. One at the very top end (albeit, using their entry level model), one from the lower end of the SLR lens market.

 

You have also upres'd the Nikon image, which is bound to have an effect. Why not just post 100% crops, rather than try to make the Nikon image the same size as the Leica one?

 

I do understand what you are trying to prove here, and as a D700 owner, I do know what they are like, but it would be a more valid test, IMHO, if you used two Leica lenses of similar age and design. AFAIK, a 50 Summicron on BOTH cameras would be a better comparator.

 

The fewer variables one introduces, the more valid the test. Given that you are testing ISO from the chip, you should try to make everything else equal.

 

I think it is pointless to prove which camera is actually the best regarding noise and trying to prove it with the same lense as this will NOT happen in real life.

 

The only thing that counts is the final results.

 

How many people will use a Leica lens on a Canon or Nikon or viceversa??? so, let's look at the final image.

 

I understand noise reduction software may be a solution.... but this means some more post production work and I do feel that it is a lot of extra work for people who have paid a lot of extra money.

 

For what have seen so far, m9 and M8 results are brilliant in "normal" situations but when using them in low light where they should excel... they fall short of what we could expect.

[lct]

As someone said above, the Sigma outresolves most of the D700 sensor at F/8, it will not introduce any variables that are very relevant to the test.

OK so you had a Summicron-M, a Summicron-R and a Leitax mount that you could have used for your comparison. But you decided not to use the Summicron-R because it is outresolved by the Sigma in your opinion, right? Now you state that the Sigma outresolves the D700 as well. So what's the point of using a lens that outresolves the body you intend to use with instead of a Summicron for both bodies?

[ho_co]

... Because we know Leica is not using any form of NR. ...

No, actually, we don't. I think whether that's so is the topic of several current threads here.

 

 

I used a 50 2.5 summarit @ f/8 and a sigma 50 1.4 @ f/8. ...

Tests done by a number of folks indicate that the Leica lenses' performance on sensor begins deteriorating by about f/5.6. I didn't do the test, and I'm glad you did; but since everyone is now backing you against the wall to find errors in your methodology, I'll just say that if I had done the test, I'd have made the comparison around f/4.

 

To do the test meaningfully, it seems to me, one needs to compare the picture chain and not just the lenses.

 

 

Quoting dpreview to justify method seems to me a mistake. I personally (yes, I'm one of the pedantic bores jackal addressed as 'jesus' above) think they're just another group who can put out great-looking (and woo-hoo! ¡interactive!) charts.

 

 

... I have explained (and linked to an article about) why I upres. ...

That's an interesting article. My gut feeling is that the author is missing some obvious element, but someone with better math knowledge would need to analyze it.

 

 

Despite all the nit-picking, I think it's great that you made the effort and have shown us your results.

[erl]

Part of Jackals remark (post 29 above) reminds me of when I used to sail competitively. Prior to a state championship sail off I observed one of my competitors polishing the hull of his boat with car polish (to reduce drag presumably). I remarked to him that he would do better to 'fix the nut on his tiller' before the race. He went looking for a spanner .

 

I did beat him, along with my unpolished hull.

 

I sometimes think we here are prone such practices when we should be concentrating on 'the nut hold the camera' to achieve maximum improvement.