Best way of creating set of cube faces

[af]

I am looking for input on the fastest and/or easiest way of creating
the six cube faces from a finished panorama.

Currently, I have them set up as six templates, but if anyone has, or
has ideas for, a fully automated workflow I'd love to hear about it.

This is for PTGui Pro.

[spgd01]

I had the same questions a few weeks back and the suggestion was to
create the six templates and then batch stitch. It takes about a min
to complete.

Hans gave me this:
What you need is just 6 project files from PTGui with preset for
generating the cubefaces. I have made a set you can download here.
http://www.panoramas.dk/panorama/FPP/Cubes24-1600-79.zip

What does the name mean? It means that this needs an Equirectangular
as source with the name 24.tif and it will make cubefaces of 1600x1600
with compression 79 which BTW is about the same as 50% in
CubicConverter.

The 24.tif in the folder is just a very small one for demo. Here what
you do: 1. copy your panorama (which should be a TIF) to the folder
and name it 24.tif The size should be around 6000-7000 pixels for 1600
cubefaces. 2. Open PTGui and open an empty window, choose the Batch
Stitcher from the Tools menu and click + and navigate to the folder.
Choose all 6 pts files and open. The conversion will start at once.
Takes only 1.30 on my old G5 PPC Mac.

The namings of the cubefaces will be 1_f.jpg, 1_r.jpg etc. This is the
default for the pano.swf. You need of course to rename the files if
you have several panos. I always name my cubefaces 1, 2, 3 etc so that
I can use a template which have these names in the XML.

If you need to convert to another size you have to open each project
file and change the output size.

Easiest is to make duplicates of the folder and edit them. You can
make sets with different cubeface sizes or compressions so that you
can easy make any size. You can also of course generate tifs for using
them to edit Nadir and Zenith. If you need to extract a nadir just use
the D.pts as atemplate.

Daniel

[spgd01]

I have a question about putting the six cubes back into an
equirectangular image. After retouching the nadir and I try to put
them back in PTGUI to one image I always get visible seams between the
cubes. However, if I use the same cubes in a flash projector they look
great with no seams. How are people converting the 6 cubes back into
one equirectangular image using PTGui. Here are my files:

http://www.sarno.biz/equi.zip

On Apr 30, 2:10 am, af <arefla...@gmail.com> wrote:

[ala...@blueyonder.co.uk]

You don't use that method to recreate the equirectangle once the zenith
and nadir have been touched, cube faces are used only to display in a
flash panorama.

To retouch zenith and nadir for a QT.mov output need a diferent technique,
there is a really good tutorial by Peter Nyfeler
http://wiki.panotools.org/Edit_zenith_and_nadir_in_one_go_with_Adjust_filter
which once is done it can be reverted to an equirectangle.

I hope this helps,

kind regards,

Rodrigo

[af]

Thanks for your replies.

Pending support for hooks to the OS in scripting, I guess batched
templates is the best one can do. It's admittedly pretty lazy as
is ;-).

@spgd01:

I also store all finished and "master" panoramas as equirectangulars.
Using the PTGui Pro viewpoint feature and editing the alpha channels
in Photoshop prior to final stitching (checking the editor for areas)
leaves only very minor touches that do not require the extraction of a
nadir. This really is preferable to patching and cloning in Photoshop.
Otherwise, the mentioned CubicConverter actually provides an easy way
of converting cube faces back to an equirectangular panorama.

But I do think that you touch upon somehting important in your follow-
up that perhaps should be investigated further. When remapping/
resammpling an extracted nadir image to be inserted back into the
equirectangular panorama, I have experienced very slight, pixel-level
mismatches in the results from PTGui. Doing the same transformations
with the Panotools plug-ins, with the same sizes, it has been spot on
in my experience. I have, however, not tested this properly since, as
described, my workflow no longer requires it.

As a sidenote, I should add that I need to produce very small cube
faces here and the downsampled result from the Bicubc and Lanczos
interpolators in PTGui is far superior to that produced by those most
relevant in the Panotools plug-ins, Spline 36 and Sinc 256. This may
be something to consider for all adjustments and sizes. Personally, I
am sticking to PTGui.

[HansNyberg]

As you mention Cubic Converter I assume you are on Mac.
In my opinion Cubic Converter is superior for ectracting converting to
cubefaces and back as long as you just use 8 bit.
Unfortunately it does not support 16 bit.

The standard bicubic in CC is just as good as the best interpolator in
Pano2VR (blackmann/sinc)
But 3-4 times faster. I never understood why.
Pano2VR is exceptionally slow and interface is awful compared to
CubicConverter.

You say you produce very small cubefaces. I would not do that from a
large equirectangular directly.
I never resample directly to smaller than around 70%
If I need smaller I resample the equirectangular first and optimize it
with unsharp mask.

Hans

[Wim Koornneef]

Hello af,

Luckily there are more roads then just one to go to Rome ;-)

My preferred road is to stitch the equirectangular in PTGui, save it
in highest quality JPEG (private use) or TIF (business use) and then
convert the equirectangular to cube faces with Pano2VR.

When the only purpose of creating the cube faces is to edit errors,
placing a nadir cap or whatever, then it is easier to create a
rectangular patch image of the area you want to edit in Pano2VR for
editing in Photoshop.

Another road I used in the past to get the cube faces is to create a
high quality compressed QTVR mov with PTGui.
The QTVR mov contains already the needed cube faces and those can be
retracted from it without any quality loss in Pano2VR.
I don't use this road anymore because for creating the QTVR mov PTGui
first internally creates an equirectangular, convert it to cube faces
and then builds a QTVR mov with the cube faces so first creating a
QTVR mov don't offer advantages(*) over the work flow in Pano2VR. (*)
The exception is the fact that the QTVR output is better to view then
a equirectangular output.

Success in choosing your road,

Wim

[HansNyberg]

I just did a test with a recent stitch.
Original 11000x5500 Conversion to movie and cubefaces at 2400x2400
Time including importing image, setting up the conversions and saving.
Mac dualcore 2,0 PPC G5

Cubic Converter; Total time 2.10. Final size 5,6 mb with standard
bicubic and compression 42%
Pano2VR: Total time 13.50. Final size 5,6 mb using Blackmann/sinc and
compression 72%

The quality is practically the same when checking a cubeface in
layers.

If you use the standard Mitchell interpolator in Pano2VR you get a
very soft unsharp result. Time is 3.10 still longer than CC.
With Lanzoes3 Pano2VR take 6.28 but still produces a less sharp
cubeface than the standard CC bicubic.

Hans

[af]

Agreed.

But sometimes one can stare oneself blind on the minute differences of
a single digit in a radius or threshold setting while time goes by.

In this case, I have developed a pano viewer for the iPhone and due to
some webkit quirks and the need for faster than fast downloads (tiny
images), there is precious little to gain through step interpolation
and all that other wizardry we like to experiment with for fullscreen
panos.

Anyone with an iPhone or iPod touch can, for instance, see the world's
largest tree (I should add, according to locals) here:
http://mobilvisning.no/iphone/mexico

Please do note that this is for Mobile Safari only. Others get a web
site - no pano.

Time, effort and output all taken into account, PTGui with Lanczos
(sharpening thus included) does a very good job.

[HansNyberg]

Ok I assume that the size you need is around 500x500 pixels cubefaces.

If you do that directly from a 6000x3000 pixels panorama without first
downsampling it in photoshop, it is very important that you choose the
best interpolators.
Cubic Converter does not have them,
PTgui only has them if you use panotools for warping.
This is what happens if you use the standard Lanczos or bicubic and
downsample to much.
http://www.panoramas.dk/panorama/interpolators/downsampling-cubefaces.jpg

The only interpolators which gives acceptable result are the 3 best,
Lanczos 3 to Blackmann/sinc
This is also the same for Pano2VR.

For automation you have in my opinion 2 options.
1. Downsample first in Photoshop with action and use CubicConverter
batch function.
2. Use Pano2VR Droplet function and one of the best interpolators. You
can use the droplet for cubefaces also but it is a little hidden.
You have to set it up in the right side as a Transformation.

Hans

[Keith Martin]

Sometime around 2/5/09 (at 09:11 -0700) HansNyberg said:

>2. Use Pano2VR Droplet function and one of the best interpolators. You
>can use the droplet for cubefaces also but it is a little hidden.
>You have to set it up in the right side as a Transformation.

Nice tip, thanks! I'd (stupidly) mixed the drang & drop feature up
with making a droplet in my mind and assumed it was only for QTVR.

k

[HansNyberg]

Actually I do not really use it myself. I always found the interface
of Pano2VR to be very difficult to use.
Its the first software I needed to use a manual to use.
Some of the functions are not what I call intuitive.

Hans

[Keith Martin]

>I always found the interface

>of Pano2VR to be very difficult to use [...]

Funny, I find CubicConverter to be no easier or harder to use, just
quite a bit different.

Both apps have aspects that could (I feel) be improved, but they're
both not *that* hard - are they?

k

[af]

I don't generally like to rely on software that is no longer under
active development. CC falls into this category imo. Pano2VR I don't
own.

Here's the image factory, aka folders, that I set up (based upon the
same idea as the one provided by Hans via spgd01 above):

1. Folders:

Root/Cube
Root/PTGui templates
Root/Photoshop droplets

2. Drag the master pano to root and rename it to a set name used in
the PTGui scripts, like pano.tif. A tip for Mac is to use the Enter
key instead of mouse events here.
3. Drag the master pano to a Photoshop droplet for resizing if
desired.
4. Open PTGui and run the templates in the batch stitcher. Set desired
face names in the scripts and save output to the Cube folder.
5. Drag the entire Cube folder (can you tell that I like OOP?) to your
Photoshop droplets: convert to sRGB 8-bit, export to format/
compression, etc. I tend to split my actions for flexibility.

While not batchable, it reduces the manual clicks and drags to a few
seconds of hard labor and employs tools already in use. Combining
different PTGui templates and a selection of PS droplets, cubes of
various sizes and with different compressions can easily be produced
to fit requirements or tweak results. Not least, each step in the
workflow itself is editable and open to changes and/or improvements.

PTGui FTW! ;-)

[Erik Krause]

af wrote:

> 5. Drag the entire Cube folder (can you tell that I like OOP?) to your
> Photoshop droplets: convert to sRGB 8-bit, export to format/
> compression, etc. I tend to split my actions for flexibility.
>
> While not batchable, it reduces the manual clicks and drags to a few
> seconds of hard labor and employs tools already in use.

Perhaps it is worth to know that you can use photoshop droplets from the
command line. Just call the droplet EXE with the images to be processed
as parameters. Hence it should be easy to automate most of the workflow
in a shell script.

There are some cube faces related scripts and templates on my page:
http://www.erik-krause.de/ttt

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

[af]

I don't have an exe since I use a Mac, but I'll certainly look into if
this can be done on a Mac. Thanks.

This one is for Joost if he's reading this thread:

When Panotools is selected as warper/stitcher, the #-interpolator
setting is not stored in the resulting PTGui project file.

The last value set with PTGui selected as the stitcher remains and is
not updated regardless of Panotools-related changes in the GUI.

Selecting PTGui as the stitcher saves changes.

This can be checked by changing the values and reading the saved
project files for each setting.

I don't know if this is by design or necessity, but it is certainly
something to be aware of when using project files.

[PTGui Support]

Hi Are,

The PTGui interpolator is stored in the #-interpolator line, and the
PTStitcher interpolator is stored in the 'm' line. So they are two
different parameters. I think it's working correctly.

Joost

[af]

Indeed and of course. You are absolutely right and it's working
correctly.

One day soon I'll discover the scrollwheel.

Thanks.

[ozbigben]

> Currently, I have them set up as six templates, but if anyone has, or
> has ideas for, a fully automated workflow I'd love to hear about it.

Why not use a single project/template file?

Load the equirectangular image 6 times
Apply template with...
* Panorama 90°, rect, square output etc...
* Yaw/pitch values as required for the cube faces
* Output LDR "Individual Layers"

This outputs 6 files, sequentially numbered.

To automate several. Open project file in a text editor and do a find
and replace on the panorama file name you want to convert. Save each
project and then create a batch.

Ben

[HansNyberg]

Ben
Problem with using 1 project is that you will need to rename every
single cubeface with the correct names.
And as they are individual it can not be batch renamed. I usually use
iView for renaming.
By using 6 projectfiles you just need to rename the original before
running it.

It is all about what gives you most work.

Hans

[ozbigben]

You can also fix the output filename and then create a simple .bat
script to rename the output files. Long live the commandline ;-)

Ben