Sigma 8-16mm used on full frame - possible

[David Kilpatrick]

During testing the new Sigma HSM II 12-24mm and the old EX DG, and also the 8-16mm, today I decided to test the 8-16mm on the Alpha 900. It does not trigger the auto crop, so the full coverage of the lens can be used (or could be used, if the petal lens hood was not built in).

Here's the full A900 frame with the lens set to 16mm, at which focal length it covers without problems from the hood:

816mm-16fullframef9acr-corr-web.jpg (313.03 KiB) Viewed 1877 times

This is at f/9 and it's by no means bad considering the intended uses. What is also interesting is that using the latest ACR in CS6 I also shot an 8mm shot, and then made the largest crop I could from this frame. The result is quite a bit wider in angle than the 8mm 1.5X sensor area - maybe equal to a 7.5mm.

And then I went further, and this is where it gets quite fascinating. I had cropped the image, and it ended up a little over 12mp, A700 size. The lens correction controls in ACR allow you to set SCALE, and zoom in on the centre of an image, blowing it up to fill the full megapixel count. So I did exactly that and zoomed in 146% so my 7.5mm view was 24 megapixels.

It seems that the deBayer calculations are used because the result does not look like a scaled Photoshop image, it holds loads of detail and could easily be mistaken for a native 24 megapixel frame. I'll write about this on the website soon.

cropscaled8mm.jpg (468.02 KiB) Viewed 1877 times

This is the result scaled down.

David

[agorabasta]

It seems that the deBayer calculations are used because the result does not look like a scaled Photoshop image, it holds loads of detail and could easily be mistaken for a native 24 megapixel frame.

Yes, I've been noticing that for quite some time that, the Lr scale control may really increase image resolution. Upscaling reduces spatial aliasing, lowers intermodulation-induced acuity and it really makes small sharply defined shapes much more recognisable.

In fact, a Bayer capture should deliver best real resolution if demosaiced to 200% linear res. That's because it's only then that the true values of gradient sitting on the green diagonals between the green sensels may actually be preserved unchanged in the output.