Pekka, any new images from your 10D and LinearSharpen?

[D30Photo]

Hi Pekka,
Looking forward to seeing more new images from your 10D.
BTW...have you had a chance to play around with LinearSharpen for the 10D or are you still waiting on Canon to fix the Linear TIFF in their conversion?

Best Regards,

[Pekka]

d30photo wrote:
Hi Pekka,
Looking forward to seeing more new images from your 10D.

OK, see below. I've done mostly casual shooting to test the camera's focusing behaviour, exposure behaviour, dynamic range limits and other things like that.

BTW...have you had a chance to play around with LinearSharpen for the 10D or are you still waiting on Canon to fix the Linear TIFF in their conversion?


LinearSharpen D60 version 1.0 will seem to handle 10D fine on color part, sharpening needs some tweaking. But for now tweaking can not be done as linear format is unusable with 10D because of the Canon's bug. So yes, I am still waiting on Canon to fix the Linear TIFF in their conversion.

Here's some 10D photos, lamp and bottle are with 35/2, horses, portrait and cactus is 70-200/2.8L, birds are 70-200 + 1.4X.

http://photography-on-the.net/10D/CRW_0245.jpg

http://photography-on-the.net/10D/CRW_0148.jpg

http://photography-on-the.net/10D/CRW_0150.jpg

http://photography-on-the.net/10D/CRW_0632.jpg

http://photography-on-the.net/10D/CRW_0813.jpg

http://photography-on-the.net/10D/CRW_0169.jpg

http://photography-on-the.net/10D/CRW_0792.jpg

http://photography-on-the.net/10D/CRW_0519.jpg

[D30Photo]

Thanks for posting the images Pekka.
The color and clarity are very good. Were these processed with LinearSharpen 1.0?
You mentioned the bug in Canon's Linear Converter only affected the sharpen method in LinearSharpen, so if I just use convert only, it should work well.

Could you post the difference between the straight-from-camera and the images you are showing to see how much improvement was made, especially on the last image of the cactus?

I had a 10D for a week, but I couldn't for the life of me get it to look the way I like. I was pretty sure it wasn't the camera because my D30 look worse using straight JPG.

I really like the color of the last image of the cactus. I just couldn't get the vibrant green that you are getting. The green I got, along with the entire image, has a yellow cast to it. I'll try to process some of those image again using LinearSharpen 1.0 from the D60 as you suggested.

BTW...other than the price, what help you decide between the 10D and the 1D? I initially went with the 10D because of the weight and price, but now I think I have to shell out the money and hit the weight room to beef up my arms since I'm seriously thinking about the 1D 'cause I got a few weddings line up this year.

Thanks again,

[Pekka]

NO, please do not use any LS now for 10D as the linear TIFF's produced by any conversion program are really 8 bit linear tiffs so you do not get a decent image out of them.

The "temporary" workflow I used for above photos is found in http://photography-on-the.net/forum/showthread.php?t=9321

[Timo Autiokari]

But for now tweaking can not be done as linear format is unusable with 10D because of the Canon's bug.

Hello Pekka, has this this bug (8-bit/c data in 16-bit/c codespace) been confirmed by Canon? If not then has somebody been actually looking at the RGB values in a linearly converted TIF, from those codes it can be easily seen if there is only scaled up 8-bit/c effectively.

Timo

[Pekka]

timo autiokari wrote:

Hello Pekka, has this this bug (8-bit/c data in 16-bit/c codespace) been confirmed by Canon? If not then has somebody been actually looking at the RGB values in a linearly converted TIF, from those codes it can be easily seen if there is only scaled up 8-bit/c effectively.

Timo


Hi Timo,

You can easily see the limited resolution by just doing curves to linear TIFF. The posterization resulted is exactly the same you get if you convert a decent 16-bit linear TIFF to 8 bits before curve manipulation.

Here is a crop from BreezeBrowser's normal conversion:

http://photography-on-the.net/chris/CRW_0034.jpg

Here is unsharpened crop of same file done from Linear TIFF with curves and saturation (colors are slightly different but that's not the point here):

http://photography-on-the.net/chris/CRW_0034L.jpg

I see no other explanation than bit truncation? I can send you a RAW to examine if you're interested. I have wrote about this bug in several occasions ( I noticed as soon as I converted first files from 10D) but not reported to Canon as I am not subscriber to their SDK - Chris Breeze knows about this but has not commented yet, I'm sure he is discussing it with Canon. I also pointed this to Phil Askey but he has not commented it either.

[Timo Autiokari]

Hello Pekka,

from those JPEGs it sure looks that way but one can not be sure from JPEGs be they edited nor unedited.

Did you also tried with the Canon ZoomBrowser (or the Viewer)? It could be a Breeze problem also.

I'd be very interested in examining the linear TIFFs in 16-bit/c mode, converted with Canon's own tool and with the BreezeBrowser. Small unedited crops, less than 1MB each, that have some dark and light areas would do nicely (my e-mail service does not pass very large files).

Remember that there is a bug in Photoshop in how it shows the image data from a 16-bit/c linear file. It first truncates to 8-bit/c, then applies the montitor profile. Should naturally happen in the reverse order. This has at least two effects:

1. You may see some banding in very dark images. If you see it then do nothing else than convert to e.g. nativePC or to sadRGB and the banding goes away.

2. When doing strong/moderate adjustment with Levels or Curves (could affect to some other tools also) the preview image (before you press the OK button) may seem to be seriously bad, weird or has banding. Then when you actually make that adjustment (by pressing the OK button) those "previvew effects" go away and you will again see a smooth edited image.

Timo

[Pekka]

Timo,

This problem is there in all converters that use Canon SDK code as base. I get identical results on Breeze and Canon's own software. You can see it in appearance very clearly, histogram peaks like a badly edited 8-bit photo and even 9MB original crop of a complex photo zips to 2MB!

This problem is also confirmed by Cathy Brown (who tried to do a profile for 10D) and others in Dpreview.

Here are some files for you:

Linear TIFF (crop). It's from Breeze (Canon's was identical in result so no point uploading it):
http://photography-on-the.net/timo/CRW_0791L_crop.zip

Original RAW from 10D:
http://photography-on-the.net/timo/CRW_0791.CRW

Simple test curve as an example:
http://photography-on-the.net/timo/simplecurve.zip
which produces result like in
http://photography-on-the.net/timo/CRW_0791L_crop.jpg

I'm glad if you can check this out in technical terms.

[D30Photo]

I'm glad I wasn't the only one to have ran into this problem. This is EXACTLY what I got when I try to use LinearTIFF (process with LinearSharpen D60 convert only). I hope it's just in the Canon's software conversion and Canon can get it fix soon.

[mskad]

No sharpening. 8MB 16-tiff LZW compression Adobe RGB color space. Didn't have any clue about how the dog looks like, so I don't know if the colors are OK. I used my D60 color table for the PowerShovel conversion.

home.attbi.com/~digital_photo/CRW_0791_PS_crop.tif (http://home.attbi.com/~digital_photo/CRW_0791_PS_crop.tif)

Hope this help.

Mskad.

[Pekka]

Thanks mskad,

That looks like a decent 16 bit conversion. Powershovel does not use Canon SDK does it?

[mskad]

Pekka wrote:
Thanks mskad,

That looks like a decent 16 bit conversion. Powershovel does not use Canon SDK does it?



It uses David Coffin's decoding algorithm and VNG interpolation.

Mskad.

[Timo Autiokari]

Hello Pekka,

thank you for the material. I've been playing several hours with the CRW but can't say anything yet, argh... but I realized that the so called 16-bit/c space of Photoshop (that truly is 15-bit + 1 level per channel) is totally useless for this kind of inspection, the mere action of opening a 16-bit/c TIFF to PS fills the lowest bits with round-off errors, even when one converts an 8-bit/c image to the so called 16-bit/c does the same. And it seems that my own histogram evaluation tool needs some rework, there are so many ways a TIFF file can be built up. Phew. So I'll be back in a couple of days, this is very interesting.

Could someone else with a higher end histogram analyzing software please also look at this?

Btw it seems that for those who do not have the 10D there is no way to convert the CRW using Canon software? I did not found anything from PowerShot nor Bebit pages. Even if it is not apparent there can be differences between those sw that use the Canon SDK, those lowest bits are not very visible natively. I converted the CRW with BreezeBrowser.

Timo

[Timo Autiokari]

Hello Pekka,

I managed to extract the image data from your CRW_0791.crw that was first converted to TIF by BreezeBrowser, it is here: http://www.aim-dtp.net/aim/temp/crw_079l_rgbrgb_dbyte_big_endian.zip about 11MB. You can verify that it is the actual data of the image by appending the filename with ".raw" and opening that to Photoshop with the following raw settings: Width=2048, Height=3072, Count=3, Interleaved, BitDepth=16, ByteOrder=Mac.

The data has odd channel values and has quite dense histrogram there in the 16-bit/c space so either:

1) something is wrong in the CRW,
2) something is wrong in the BreezeBrowser,
3) these odd values and dense 16-bit/c histogram are the result of noise reduction in 16-bit/c space inside the camera (or in the conversion sw, Canon SDK).

Since the 12-bit raw sensor data is scaled up to fill the 16-bit space the up-scaled data should only contain even values that are equally spaced, 16 levels apart (in case if we have true raw data).

Now, in case of (3) there will not be a bug fix I'm afraid.

If you examine the converted TIFF with my Bit-Slice filter using Binary mode you can see what kind of variations there are in the different bit-planes (slice 1 means LSB or bit 0). The LSB is extraterrestrially smooth and slices 2 to 7 are so too, then the slice 8 (bit 7) is almost pure noise and there on the higher planes get gradually smoother and smoother. From this I think that the bits 0 to 6 are artificial (result of noise reduction etc) leaving bits 7 to 15 (yes 15 since internally Photoshop uses only 1 level from the 16th bit) for real data from the sensor, from which one bit is almost totally noise. So this gives something like 8 bit effetive image data.

I sure hope that I'm totally wrong and and that it eventually turns to the best. But it is absolutely sure that the TIF from the BreezeBrowser is not just a 8-bit/c data that is scaled to 16-bit/c space.

Timo

[Pekka]

Thanks Timo,

I sure hope that I'm totally wrong and and that it eventually turns to the best. But it is absolutely sure that the TIF from the BreezeBrowser is not just a 8-bit/c data that is scaled to 16-bit/c space.

I think it is 16 bit data which is built from 8 bit image. :)

I'll point this discussion to Chris Breeze and wait for his comments.

[Mickey]

Hey Pekka,

I just ran that RAW through my Powershovel workflow and the colours turned out very well (IMO). I'd be happy to send you an email with a full sized JPEG if you're interested (about 1.42MB). I also tried running LinearSharpen on the linear Powershovel output. I know it's not designed to handle these files so the result is rather irrelevant, but it looked like it had too much contrast for my liking. The sharpness looked to be pretty similar except using High sharpening tended to emphasize noise in the bokeh more than my normal method. I look forward to trying your LinearSharpen once Canon fixes the 16-bit bug.

If you haven't yet tried Powershovel 2, here's a link to the latest version I just put online this morning in case you're interested: http://www.morpheusmultimedia.com/ps/ps2.html

Michael

[Chris Breeze]

BreezeBrowser uses the same Canon raw conversion libraries for 10D conversion as Canon's File Viewer Utility. I've just compared the output from BB v2.6 and FVU 1.2 with 10D 16-bit linear conversion and the results are identical.
It appears to be a bug in the Canon raw conversion libraries, not in the way BB handles the image data after conversion. I've reported the bug to Canon.

Regards,

Chris