The 21MP full RAW file contains much more data/resolution than the sRAW file. So, it is only logical that the downsized full RAW file appears sharper than the sRAW file.
I can't think of any place for user error. It was tripoded, the focus wasn't touched, there was no wind and 1/640 shutter speed. I tried it a few times with similar results.
It looks the same in DPP (not that I care, I'm not going to stop using LR).
Ha. Brent... I was going to say you were crazy but I actually think you are right.
1. Camera Tripod Mounted
2. Exposure setting the same
3. Live View Manual Focus so no focus change
4. Opened both in ACR default values
5. Image resized Full Raw to size of sRAW in Photoshop
6. Cropped to 100%
* focus on "auxilios"
I tried it when I first got the camera's and came to the conclusion that RAW resized down to SRAW or any other dimension in PP was clearly sharper. I don't believe that whatever formula Canon is using in-camera delivers the same results as taking the 21mp image and doing it yourself.
Actually now that I think about it, it does make sense and its one of the reasons why sRAW has never appealed to me. With sRAW, the camera takes the RAW image, demosaics it, then downsamples it and saves linear demosaiced pixels in the "sRAW" file. The final sharpness is going to be heavily influenced by the choice of on camera downsampling algorithm and I think thats what we are seeing here.
If it demosaics it and then saves the new data structure, shouldn't we be able to "see" it after correction for linear mode and WB? IOW it's already partly converted. Do we it's demosaiced or is it just sampling the "normal" readout data and selectively storing values as if it were a smaller sensor saving a normal raw? I'm not up on white papers recently so sorry if it's already known.
Thanks
Andy
It probably means Canon improved whatever method they were using to downsize the RAW file. The difference in the 5D is so small, for me, I am not worrying about. Just a little extra sharpening in post...
Interpolation has to occur in order to produce the sRAW file. It follows that the file you get will not be as crisp on the pixel level as a native resolution file.
I have noticed the same thing, but the actual drop in sharpness is very small. I don't consider it a big deal for most things.
I have noticed this too, though in normal use it's not a big deal. The files sharpen just fine in post.
The 1Ds3 sRAW resolution is a lot lower than the 5D2 sRAW1, so perhaps Canon is using a pixel binning algorithm in that camera that is unavailable to them for the 5D2 sRAW1 implementation. I think the 5D2 implementation is a lot more usable in terms of resolution.
I use sRAW1 for event work and volume stuff where the client does not need a big file and I'm sure I can live with 10Mp. Files are sharp enough for that.
I "black boxed" sRaw in 5D2 a while ago, and it seems that it deals with colour noise in a slightly different manner than the normal in-camera raw-interpolation. At least as far as I have been able to tell from my calculations from identical sRAW/downsized raw comparisons... This might be that the raw-conversion uses a bigger area for colour estimation, and that might affect perceived "sharpness", but it's not an effect that I've noticed. I did all the comparisons at ISO3200+ though, so this isn't really the same thing. Noise affects sharpness/detail quite a bit - and in noisier pictures than the examples shown here I found no sharpness difference. Only a slightly higher medium frequency contrast, and less (and slightly differently spaced) colour noise.
When it comes to sRaw2, the interpolation is 2:1, so I'd guess that Canon has chosen to
1 - use a bigger area for colour estimation in the raw-interpolation
2 - use 4:1 averaging when downsampling (like binning/averaging. Take the average of four > make one pixel). This is a lot less computationally intensive (less work for the processor in the camera), and actually a lot sharper and more accurate than normal bicubic. The sRaw2 is sharper than downsizing a "normal" raw-conversion to 50% with bicubic, if you measure it by imatest.
My conclusion then was that sRaw was so-so usable, but sRaw2 gave some subtle advantages - when compared to resampling a normal raw. So if you know that you're going to use a picture at <5MP anyway, sRaw2 is actually quite good.
