Hi,
 
Can anyone tell me the default settings, or preferably, the best settings for avoiding any kind of jpg compression in stitcher 5.5.4, on a mac book pro OSx 10.4.11
 
I am still experiencing posterization, or banding, particularly evident in the sky! I have posted other threads on the forum, but believe, due to the expert feedback people have offered, that it is down to jpg compression. I have tried converting my raw files to 16bit .tif files but am unable to render them. I still experience some banding when i render out at the following settings:
 
Images shot on Canon 5D, with Canon 15mm Fisheye, at LRG JPG:
 
- Spherical image - 100%
- TIFF
- Blending Method: Smart
- Sharpen: Maximum
- interpolation: Lanczos5
 
Thanks,
 
Chuck

Chuck...
 
You might check your "QT OUTPUT SETUP"...
I'm getting good results with these settings (even in blue sky) when creating a Cubic QTVR from a Spherical:
 
Render Properties: Morph  Normal  Lanczos5
Compression Settings: Codec - Photo/JPEG
                               Quality - 50%
                               Rendering Quality - Motion/50%  Static/75%
                               Tiling - 5x5  Auto Tiling 400 pixels
 
Todd

Todd,
What do you mean by "QT OUTPUT SETUP"?
Are these settings within stitcher or within quicktime?
 
Chuck

Hi Chuck!
 
Instead of trying to render 16-bit TIFFs why don't you try 8-bit - perhaps you are hitting a RAM limitation in your renders. How much physical memory do you currently have? If you use the Mac's activity "Activity Monitor" (look in the "Applications" > "Utilities" folder) you can track what Stitcher is doing with the RAM.
 
I believe Todd is referring to a QTVR render in Stitcher with his reference to "QT Output Setup" - you will not find this dialog with a spherical render.

Hi Jim,
 
I'm using a MacBook Pro with 2gb of RAM. I've converted my raw files to 8 bit TIFFs, stitched them and rendered at :
 
Spherical - tif - 78% (9000 x 4500 approx)  
Blending Method - Morph
Sharpen - Maximum
Interpolation - Lanczos5
 
&
 
Spherical - tif - 78% (9000 x 4500 approx)  
Blending Method - External (Enblend v1.3)
Sharpen - Maximum
Interpolation - Lanczos5
 
&
 
Spherical - tif - 78% (9000 x 4500 approx)  
Blending Method - External (Enblend v1.3)
Sharpen - Normal
Interpolation - Bicubic
 
The result is far better using the external blending method, however it still appears as tho there is some posterization on the sky. Strangely, the final test using a Bicubic interpolation yielded the best result, i/e. the least amount of posterization in the sky! If I were to post some images online, would you be able to have a go at processing them to see what result you get?
 
Jim, would you be able to explain some of the render settings to me as I'm a little confused.
Originally I had been using the following settings in stitcher unlimited 5.5.4:
 
Blending Method - Morph
Sharpen - Maximum
Interpolation - Lanczos5
 
Then I read the following post about a file missing from stitcher unlimited - http://forum.realviz.com/realviz/D [...] 2745_1.htm
 
What do each of the above settings actually do? How does the interpolation setting effect the panorama? and should I be opting to blend through an external application, such as enblend? Is this because stitcher's blending methods apply's some/more compression?
 
Many thanks,
 
Chuck

Chuck,
 
Let me clarify the process I go through…
 
I do all of my work in Stitcher and use Photoshop to clean things up.
After I’m done stitching my images, I render them out as a Cubical Image / Photoshop (PSD) at 100%... Morph / Normal / Bicubic / Use Blending…  (this process takes a while depending on how fast your computer is)
 
I then “clean up” the six different sides (files) of the cube, removing any blurring, etc in Photoshop…
 
I then “Load Panorama”, importing the (6) Photoshop files back into “Stitcher”…
 
I then render out a “Spherical Image” TIF at 100%... Morph / Normal / Bicubic…
 
I take the spherical TIF file into Photoshop and make any additional retouches that are needed and resize the image to 6000 x 3000 pixels and save…
 
I then “Load Panorama” (smaller TIF) and render it out as a “Cubic QTVR”…
1500 x 1500  / 100% … Morph / Normal / Lanczos5… Photo – JPEG / 50%...
Motion / 50%... Static / 50%... 5x5 / Auto-Tiling / 400 pixels… Window / 1064 x 600…
 
I can send you a sample of my finished QTVRs to see how they look on your computer…
Please contact me at    tsalmon@lesea.com and I will forward the files to you…
(I don’t know of a way to post them on this site)
 
Hope this helps!
Todd
 

Hi Chuck!
 
Right now I'm on a very tight deadline for presentation tomorrow (Sat 1/12) - Saturday night I will be able to reply in-depth to your questions.
 
"If I were to post some images online, would you be able to have a go at processing them to see what result you get?"
 
Definitely - but they need to be TIFF from RAW source images - JPEG is a nonstarter at this point in the process. 8-bit TIFFs might possibly work, we'll see (but we don't have much choice since 16-bit doesn't work for your set-up); and since the sky banding is the problem we are trying to solve here you do not need to include the bottom cap shot(s).
 
-----------------------------------------------
Also something that just became apparent to me as I was looking over your recent methodology - you should not be using sharpening in your spherical renders from Stitcher. Do this operation in Photoshop. Stitcher's sharpening is OK if you are going straight from stitch to QTVR. But sharpening in Photoshop is far better for processing a spherical because you can use a variety of sharpening techniques and apply them selectively to the spherical image. Applying sharpening in the Stitcher render is a shotgun approach (and not overly refined) - there is no reason to sharpen a blue sky for instance - it will only enhance any banding tendencies - or other artifacts, and it reduces the over all quality of the image (removes data).
 
Sharpening is the last thing you should do in a spherical workflow - just before the final QTVR render.
 
 
I highly recommend DXO for superior imaging results (but beware Windows users - the most recent release of DXO v5 has been quite a mess to date. Mac users can still get the v4.x version - excellent!). If DXO supports your camera body/lens combo it "deblurs" the inherent flaws in the images. This is different from sharpening the image.
 
Please see these comparison conversions of a RAW from Photoshop CS2 and a RAW from DXO here.
 
No sharpening used in the RAW conversion for either image. Both programs are on their default settings, with the exception of the "deblur" control in DXO.
 
Because you start out with a much more "focused" image with DxO you do not need to lay on such a heavy sharpening routine in the final spherical image (just before conversion to QTVR); beneficial for the image quality of the QTVR and improves its compression.

Hi Chuck!
 
RE:
Blending Method - Morph  
Sharpen - Maximum  
Interpolation - Lanczos5  
 
 
The manual actually does a good of explaining the Blending & Interpolation options (see page 90 in the printed manual for Blending info & page 95 for Interpolation).
 
Blending
The manual is too brief though on External blending - but this is the option you will find the most valuable for producing spherical renders. I do not understand the math or actual image processing technology behind the blending programs (though check out Stitcher v5.5 and run enblend and watch the processes run in the Terminal window - this window does not appear in v5.6), but basically these programs analyze the VR and evaluate where in the VR overlap zones are errors/mismatches and corrects these flaws - everything from slightly skewed flag poles to mismatched tones in the sky.
 
But these blending programs also do something else - if they cannot fix an error they often create sharp breaks in these locations. Usually these sharp delineations make for an easy Photoshop fix. This is different than the "Morph" blending option (also tries to make the overlap zones match through "warping" ) which will give you a blurry area where it cannot fix things. These blurry zones are often much harder to correct in Photoshop.
 
For an example of errors that enblend cannot fix - and the resulting appearance please see:
http://web.mac.com/wdd3/iWeb/Laser [...] pod-2.html
This is a handheld I did and purposefully did not fix the flaws. The most flagrant one is the umbrella that is about 90 degrees to your left from where the VR opens.
 
 
Sharpening
Probably have said enough about this in my previous post - but there is another philosophy on sharpening that posits you get the best results by sharpening at each stage in the process. "Pixel Genius" makes a plug-in for Photoshop that works on this principle, and I use it on images that I print. I haven't found the effort worth it for VRs - especially since I've been using DXO.
 
So the steps for a QTVR would be:
1) Open source image - apply a slight amount of sharpening.
2) Stitch & render as usual (but not using Stitcher's sharpening routine)
3) Make any needed corrections in PS then do your main sharpening.
4) Create your QTVR - or
5) If you are going to resize your spherical or cubic from the original format to create a new version, add another round of slight sharpening if needed after resizing. This can be valuable if the resizing really did lessen the quality of things.
 
I do not find Step #1 to be that effective for QTVR purposes, #3 typically is the only time I sharpen.
Experiment to see what works best on your images.
 
 
Interpolation
Basically interpolation is an operation that adjusts the value of pixels when they have been "acted" upon, such resizing an image or remapping it. This adjustment to the pixel is made based on the values of its neighboring pixels.
 
The order of these interpolation choices in Stitcher's render dialog goes from least accurate at the top "Nearest" to most accurate at the bottom "Lanczos". In "Nearest" the value of a pixel is based on its neighbor pixel. Lanczos 5 utilizes values from a 10x10 grid around the pixel.
 
Many people feel that Lanczos 5 yields the best results. I can hardly see a difference between Bicubic & Lanczos in most of my VRs, and Bicubic runs much faster (less calculations; it is a 4x4 array), so this is what I like to use.
 
As with sharpening, experiment to see what works the best for you
 
 
Look forward to seeing your images if that's something you still care to do.

Jim, you are the man!......
 
It definitely seems to be the sharpening which is having the greatest negative effect. Shooting on SML JPG obviously has it's limitations and I'll definitely be thinking about shooting in RAW in the future. I'm now finding that I can use SML (8-bit) JPG images, stitch them and render them out as 100% JPG, on 'Smart' blend, 'None' sharpen & 'Lanczos5' interpolation and the banding is drastically reduced. Obviously, the initial compression of the image when captured results in some posterization (patchy skies), but when i had the sharpening set to 'maximum' it was creating these thin, faint, white bands which were far more apparent.
 
Todd, thanks for your input here as well. I will be looking into the most appropriate method for processing my images in the future. Your work flow has raised some interesting options for me, thanks.
 
Cheers gents,
 
Chuck

Hi Chuck!
 
Reading my last post to you I realized I said something in the "Sharpening" section that could be misinterpreted.
 
In Step 5 where I say "If you are gong to resize your spherical..." I am not refering to resizing for QTVR purposes - this is best left to the render process and derived from the original-sized spherical.
 
But sometimes you may wish to reduce the spherical itself - possibly for printing; my usual reason for doing so is for displaying the (reduced in size) spherical on my website.