Using Raw to stitch - bad idea?

[Hans]

I never understood the idea of using the raw files directly in PTgui.
From what I understand there is no way of correcting CA in PTGui and
that is to me enough not to use it.

After seeing the last post by Allen Graves I got another suspicion.
He uses Raw NEF from fullframe 10.5mm and I could see that the
controlpoint generation was not at all sufficient.
In my experience Fullframe is superior for panoramas with
controlpoints generated all along the image almost always.

So I did a test on a panorama I just made using 16bit Tiff and a 15mm
on a 5D.

I loaded the same images and generated automatic controlpoints.
The result talk for them self. Top Raw file. Bottom 16bit tif
http://www.panoramas.dk/panorama/ptgui/Rawortiff.jpg

I repeated it with different setups both using the simple auto setup
and a manual and both fullframe and circular crop.
Same result. Only one of the image pairs generated useful
controlpoints.

Hans

[Erik Krause]

Hans wrote:

> I loaded the same images and generated automatic controlpoints.
> The result talk for them self. Top Raw file. Bottom 16bit tif
> http://www.panoramas.dk/panorama/ptgui/Rawortiff.jpg

Astounding! I never was a friend of using RAW files directly. However,
PTGui doesn't actually load RAW files, it loads dcraw output. The
paramters passed to dcraw are -c -4 -o 1 -w which means standard output,
16 bit linear, sRGB colorspace, and camera white balance.

I guess the linear (not gamma corrected) image data is the reason...

BTW.: There is a possibility to have dcraw remove CA: The -C parameter
together with two correction values for red and blue channel. If only
PTGui would let us specify custom dcraw parameters like for other
external helpers...

best regards
--
Erik Krause
http://www.erik-krause.de

[Allen Graves]

I stitched the 21 images using Raw, then converted the images to Tiff
and stitched again. My results below show that there is a similar
distribution of control points, but the error is greater from the Tiff
files. Also the color balance has shifted a bit in the Tiff images.

Raw: http://i17.photobucket.com/albums/b91/abgraves/raw.jpg
Tiff: http://i17.photobucket.com/albums/b91/abgraves/tif.jpg

[Erik Krause]

Allen Graves wrote:

> I stitched the 21 images using Raw, then converted the images to Tiff
> and stitched again. My results below show that there is a similar
> distribution of control points, but the error is greater from the
> Tiff files. Also the color balance has shifted a bit in the Tiff
> images.

Pay attention to the image size. dcraw does not always provide the same
image size as commercial raw converters do. dcraw apparently tries to
convert as many pixels as possible while ACR f.e. relies on some meta
data. I've seen a difference of up to 30 pixels.

If you use RAW images to find control points and then replace the RAWs
by TIFFs the control points won't match the features anymore since they
are scaled to the new size.

[Allen Graves]

Erik Krause wrote:

> If you use RAW images to find control points and then replace the RAWs
> by TIFFs the control points won't match the features anymore since they
> are scaled to the new size.

I used two completely different projects: One exclusively for the Raw
images and one exclusively for the Tiff files. The results were just
not dramatically different either way.

So do you shoot Raw with your camera and convert to Tiff before you
stitch, or are you shooting Tiff directly with your camera?

Thanks,
Allen

[HansNyberg]

I do not think there are any cameras today that uses TIF directly in
camera.
I shoot Raw then convert to usually 16 bit Tif and in most cases if I
need more dynamic range than my single Raw can give me I enfuse the 3
bracketed shots 2EV+ = 2EV- before I stitch.
I remove all CA and vignetting in Phoptoshop RAW during the
conversion.

I have done some more experiments with RAW and TIFs and I found that
exposure is actually what affects the Autocontrolpoints for the RAW
images.
This does not happen for the TIF images.

I usually shoot either without bracketing and 1/2 to 1EV underexposed.
This gives me enough to get everything I need from my 5D images in
most outdoor panos.
I also do the same for bracketed as I only have 3 exposures max +-2EV
on Canon to play with.

It turns out that PTGui has no problem generating lots of
Autocontrolpoints even from my 2 EV (actually they are 3EV)
underexposed TIF images.
But even from the medium exposure which is almost 1 EV underexposed it
has problems. Using the overexposed Raw images gives me the same as I
get from the TIF.

This also means that if I use linked images in PTGui I get very few
controlpoints often just in centre. If I unlink them the overexposed
images produces a lot of controlpoints.

This might be a processor thing as it uses a lot of extra time for the
Raw images or it might also be a specific problem with Canon CR2.
I just tested a set of NEF which turned out to work much better.

Next is to test the underexposed NEF, have not done that yet.

Hans