I've seen this a lot before: resizing a 32-bit file in Photoshop CS2 with their bicubic interpolator produces artifacts that look like "excessive sharpening halos" around high contrast details. Because of this, I simply stopped scaling images in Photoshop and instead use Photosphere (no problems).
However, I was recently perusing the Adobe user-to-user forums, and it was determined that the "sharpening artifacts" are only present if you subsequently save the file in Radiance format. It was then proposed by an Adobe employee that "the radiance file format code may have some problems dealing with extreme values." (keep in mind these are user-to-user forums, not developer comments).
Here is the thread in question:
http://www.adobeforums.com/cgi-bin/webx/.3bbe9c64
Now I know that Photoshop's Bicubic does some sharpening (even without using "Bicubic Sharper"), and that sharpening of any kind on an HDR file results in extreme values, but does the Radiance file format itself have a problem with extreme values? Or is this a combo of PS Bicubic and their Radiance I/O? I pretty much store all of my "in progress" files in Radiance format and then create "final HDR files" from them in EXR, so this has effected me and my workflow.
-Mark
BTW, I've also found a huge flaw in Photoshop CS2's "Merge to HDR" when small, bright details are moving between shots (think ghosted leaves in a mild breeze) - so I don't even use PS CS2 for *creating* HDRs anymore, but the PS CS2 "clone tool" and the plugin Flexify 2 are pretty much indispensable for retouching HDRs, so I'm using Photoshop in any case.
Hi Mark,
The Radiance RGBE format is not able to record negative pixel values, so such primary values should be clamped to zero. Apparently, Photoshop is not doing this. Photosphere (and the Radiance tools) check for and truncate negative primaries, writing them out as 0's. OpenEXR doesn't have this issue, as it can read and write negative primary values -- in fact, you are better off using OpenEXR for your intermediates also as it is a higher-precision format. It doesn't really make sense to use a lower-precision format like RGBE for intermediate results then write out your final as OpenEXR, as you've lost information along the way. It would make more sense the other way around.
-Greg
···
From: "Mark Banas (lists)" <listmail@mab3d.com>
Date: June 15, 2006 2:10:26 PM BDT
I've seen this a lot before: resizing a 32-bit file in Photoshop CS2 with their bicubic interpolator produces artifacts that look like "excessive sharpening halos" around high contrast details. Because of this, I simply stopped scaling images in Photoshop and instead use Photosphere (no problems).
However, I was recently perusing the Adobe user-to-user forums, and it was determined that the "sharpening artifacts" are only present if you subsequently save the file in Radiance format. It was then proposed by an Adobe employee that "the radiance file format code may have some problems dealing with extreme values." (keep in mind these are user-to-user forums, not developer comments).
Here is the thread in question:
http://www.adobeforums.com/cgi-bin/webx/.3bbe9c64
Now I know that Photoshop's Bicubic does some sharpening (even without using "Bicubic Sharper"), and that sharpening of any kind on an HDR file results in extreme values, but does the Radiance file format itself have a problem with extreme values? Or is this a combo of PS Bicubic and their Radiance I/O? I pretty much store all of my "in progress" files in Radiance format and then create "final HDR files" from them in EXR, so this has effected me and my workflow.
-Mark
BTW, I've also found a huge flaw in Photoshop CS2's "Merge to HDR" when small, bright details are moving between shots (think ghosted leaves in a mild breeze) - so I don't even use PS CS2 for *creating* HDRs anymore, but the PS CS2 "clone tool" and the plugin Flexify 2 are pretty much indispensable for retouching HDRs, so I'm using Photoshop in any case.
This is because of the sharpening aspect of bicubic kernels (they really
don't work for HDR images), and can be made worse in the Radiance file
format because we weren't clamping negative values correctly.
Chris
···
On 6/15/06 6:10 AM, "Mark Banas (lists)" <listmail@mab3d.com> wrote:
I've seen this a lot before: resizing a 32-bit file in Photoshop CS2
with their bicubic interpolator produces artifacts that look like
"excessive sharpening halos" around high contrast details. Because of
this, I simply stopped scaling images in Photoshop and instead use
Photosphere (no problems).
However, I was recently perusing the Adobe user-to-user forums, and
it was determined that the "sharpening artifacts" are only present if
you subsequently save the file in Radiance format. It was then
proposed by an Adobe employee that "the radiance file format code may
have some problems dealing with extreme values." (keep in mind these
are user-to-user forums, not developer comments).
Here is the thread in question:
http://www.adobeforums.com/cgi-bin/webx/.3bbe9c64
Now I know that Photoshop's Bicubic does some sharpening (even
without using "Bicubic Sharper"), and that sharpening of any kind on
an HDR file results in extreme values, but does the Radiance file
format itself have a problem with extreme values? Or is this a combo
of PS Bicubic and their Radiance I/O? I pretty much store all of my
"in progress" files in Radiance format and then create "final HDR
files" from them in EXR, so this has effected me and my workflow.
-Mark
BTW, I've also found a huge flaw in Photoshop CS2's "Merge to HDR"
when small, bright details are moving between shots (think ghosted
leaves in a mild breeze) - so I don't even use PS CS2 for *creating*
HDRs anymore, but the PS CS2 "clone tool" and the plugin Flexify 2
are pretty much indispensable for retouching HDRs, so I'm using
Photoshop in any case.
_______________________________________________
HDRI mailing list
HDRI@radiance-online.org
http://www.radiance-online.org/mailman/listinfo/hdri
Hi Greg,
Aarrgh... right you are. A very long time ago I set up a workflow that was all Radiance file based because: 1) my 3D software only understood that file type, and 2) I read "somewhere" that Radiance format could store a much greater range than OpenEXR. Despite reading many other details about both formats, I stuck with my initial, oversimplified understanding which was "Radiance holds more brightness than EXR."
Basically, I should've read your HDR Image Encodings page even more carefully - not only is the precision in OpenEXR much higher, but most of the extra brightness latitude in RGBE is essentially unused, even for my very-wide-range "light probe" panoramas. EXR is very widely accepted now, and I think 10.7 orders of magnitude is enough for my purposes... 
Good thing I posted what I did... I doubt I'll forget it now.
-Mark
···
On Jun 15, 2006, at 2:39 PM, Gregory J. Ward wrote:
Hi Mark,
The Radiance RGBE format is not able to record negative pixel values, so such primary values should be clamped to zero. Apparently, Photoshop is not doing this. Photosphere (and the Radiance tools) check for and truncate negative primaries, writing them out as 0's. OpenEXR doesn't have this issue, as it can read and write negative primary values -- in fact, you are better off using OpenEXR for your intermediates also as it is a higher-precision format. It doesn't really make sense to use a lower-precision format like RGBE for intermediate results then write out your final as OpenEXR, as you've lost information along the way. It would make more sense the other way around
Hi Chris,
Glad you are on this list as well!
Hot on the heels of my previous post about the Radiance files from Photoshop CS2, I've put up a page to explain the amplified artifacts I have seen in "Merge to HDR" in CS2 (and why I stopped using it). Perhaps you can take a look at this and tell me where I've either gone wrong or perhaps why I'm seeing these dramatic effects. The odd part is that I don't remember having this problem in the past, but I know that I recently upgraded to 9.0.1 and because of this problem I have switched to batch HDR-creation with Photomatix instead (despite using CS2 in my workflow right *after* creating the HDRs).
Here is a page (with links to the "originals") illustrating the problem I am seeing with "ghosts" in PS CS2 Merge to HDR:
http://www.mab3d.com/temp/ot3d/HDartifacts.html
At the same time, this is a general query to see if anyone else has seen/ had these problems with Merge to HDR in CS2, and what they have discovered.
-Mark
BTW, for those that are interested, the ZIP archive of stitched exposures includes JPEGs that have the metadata from the pre-stitch images appended to them by EXIFtool - makes HDR creation a bit easier. Plus, if you peruse this data in a capable image browser you can get a good idea of how I process my RAWs for HDR creation.
···
On Jun 16, 2006, at 12:57 AM, Chris Cox wrote:
This is because of the sharpening aspect of bicubic kernels (they really
don't work for HDR images), and can be made worse in the Radiance file
format because we weren't clamping negative values correctly.
