A6000 test cool extra - Clear Zoom Image E 55-210 vs D5300 55-300 VR exelent :-)

Started Mar 26, 2014 | Discussions thread
forpetessake Veteran Member • Posts: 4,892
Re: correcting misunderstanding

blue_skies wrote:

forpetessake wrote:

blue_skies wrote:

forpetessake wrote:

Electrona wrote:

Hello again , this is a short comparison of the Clear Zoom function of A6000 with E 55-210 OSS and

the NIKON 55-300 VR . Clear zoom is actually a smart zoom up to 2 /Zoom range 0-2 times/times and by processing the image is creating a file again 24Mp

There is nothing particularly smart about 'clear zoom', it's nothing more than a glorified crop and interpolation. One can do the same in any PP with likely better results. The only smart thing about it is the marketing, taking a pedestrian processing and presenting it as something superior.

but now we have an 'artificial' zoom 55-210 4.5-6.3 + the expanded range of 210-420 4.5-6.3

This is another error. As a result of 2x cropping, the lens becomes 110-420mm f/9-12.6 in APS-C equivalent. You get a lot less light (thus noise) and a lot deeper DOF than from the Nikon lens.

Bottom line, it's hard and expensive to produce high resolution, bright telephoto lenses, and cropping is no substitute.

I see a long debate here regarding optical zoom, digital zoom, and intelligent interpolation and upsampling methods, such as bicubic methods.

Yes, but you are repeating the same debunked beliefs.

The key part in Sony's CIZ comes in two ways: higher resolution in the A6000 at 24Mp and pattern-recognition algorithms (heuristic), rather than an intelligent interpolation.

This is just a gobbledygook. Interpolation is just that interpolation: algorithm for adding the missing information.

By upsampling 2x, the CIZ algorithm has to add a pixel for every existing pixel,

You are forgetting that 2x upscaling means 4x more pixels: three pixels out of four are added.

but it cannot average the values, for it would be similar to digital zoom. Even intelligent algorithms are subject to whichever match function guides them.

Sony otoh explained the CIZ (in past) simpler this way: if shapes are random, then any upsampling method can be accepted, as long as it maintains clarity. But if shapes are regular, understanding the regularity leads to much better interpretation of upsampling methods.

This is again a gobbledygook and not a technical discussion. All interpolation methods have certain models they implement, the same bicubic can be using Lagrange polynomials, cubic splines, or cubic convolution, and that works extremely well for any as you say 'regular shapes'. It's for random, or better say, non-smooth datapoints, that interpolation gives random results. What follows is that the bigger the crop factor the more every interpolation method looks alike. There is a way to improve guessing 1 additional datapoint for every 2 real datapoints, but there is no good way of guessing 3 additional datapoints for each real one.

I believe that CIZ applies pattern recognition and uses a (large?) set of templates. Depending on the experiment (shape and resolution) you may find that CIZ is nearly as good as optical zoom, as well as the opposite - CIZ barely outdoes digital zoom.

But do try for yourself.

And every comparison shows exactly what was shown in this thread above: CIZ is no better than the standard bicubic interpolation.

I would say that, based on my limited experience, the CIZ gets you a good 75% of the way to an optical zoom, which I find very impressive. Upscaling algorithms are NOT as good as CIZ, unless you have very high end ($$$) post production tools that specialize here (which most of us don't).

The benefit of CIZ is that it maintains your lens speed, so you don't pay for the longer (optical) lens by having a smaller aperture.

No it doesn't. Just like any cropping or a rear teleconverter it reduces the effective f-stop proportionally to the crop factor, meaning more noise and deeper DOF compared to a lens with equivalent focal length and the same f-stop. The only known way of preserving the f-stop of the lens (light and DOF) is using a front mounted teleconverter (telescope).

I don't like when people simply ignore every cogent argument and repeating the same fallacies as if nothing has been already explained.

Like I said, I 'guesstimate' that CIZ is about 75% towards optical zoom (being 100% and digital zoom being 0%).

And those numbers are of the wall? How did you arrive to them? Measured something?

CIZ is dramatically better than digital zoom, and, afaik, also more successful in recognizing certain patterns as the upsampling tools. But again, this can be hit or miss.

Just call it a woodoo zoom, because that's what you seem to believe it is.

As to gobbledygook and debunked beliefs, I am commenting on my personal results with CIZ, not some long technical debate here. I find CIZ adequate and useable, within its limitations. The 75% numbers comes from empirical results, not some special study or algorithm analysis.

Again, how did you arrive to that number?

Like one poster said, we should post pictures here and skip over the discussions.

It reminds me all those scam artists, who insist one has try what they sell instead of asking for a logical explanation.

CIZ is better than one thinks - but I should add that I don't always is 2.0x, but could be between 1x and 2x - it also changes the result.

As to f-stop - since CIZ is applied post exposure, the lens is used as is. There is no effect on aperture/speed from a CIZ perspective. I compare that with a longer lens, at one f-stop less and one stop slower speed. CIZ does NOT require this.

You simply don't understand the effect of cropping. Do you believe that rear mounted teleconverters also don't have an effect on the total light gathered by the lens? Here is a hint: the cropping and teleconverters are very much alike.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow