Hello everybody,
When I convert equirectangular to cubes I notice major quality loss. Source is 16bit TIFF and I convert to TIFF as well. What could be the reason for that? Any clues?
Cubes convertion quality loss
- 360Texas
- Moderator
- Posts: 3684
- Joined: Sat Sep 09, 2006 6:06 pm
- Location: Fort Worth, Texas USA
- Contact:
Hello TamaiT, welcome to the forum.
Working in 16bit is best.
Question? What is your stitched panorama overall width? Knowing that value will help us determine what the cube face size should be used to create cube face tiles.
Let us assume that your panorama is 6000 pixels wide x 3000 pixels high. Optimizing a cube face tile side = 6000 pixels wide / Pi or 6000 / 3.1415 = cube face size 1909.92 or use 1910
Use 1910 pixels to cut your panorama into 6 cube face TIFF tiles. The larger the cube face tile the higher the quality using TIF output.
Hope that helps
Working in 16bit is best.
Question? What is your stitched panorama overall width? Knowing that value will help us determine what the cube face size should be used to create cube face tiles.
Let us assume that your panorama is 6000 pixels wide x 3000 pixels high. Optimizing a cube face tile side = 6000 pixels wide / Pi or 6000 / 3.1415 = cube face size 1909.92 or use 1910
Use 1910 pixels to cut your panorama into 6 cube face TIFF tiles. The larger the cube face tile the higher the quality using TIF output.
Hope that helps
Hello TamaiT
From the examples in your post it seems you are actually showing the result of two transformations - the first from equirectangular to cube faces and the second from cube faces back to equirectangular. With any transformation involving remapping of pixels you can expect some degradation, and I would be surprised if it were possible to do these two transformations without a visible loss in quality.
Presumably the loss at the cube face stage is somewhat less than that shown in your post? If so, the question is 'Is that quality loss more than one would expect?' I'm afraid I don't have the answer to that.
Just a thought though - which option are you selecting for 'Interpolation filter' in Pano2VR's preferences, and have you tried other settings there to seek an improvement?
Regards
Andrew
p.s. Is that Edinburgh?
From the examples in your post it seems you are actually showing the result of two transformations - the first from equirectangular to cube faces and the second from cube faces back to equirectangular. With any transformation involving remapping of pixels you can expect some degradation, and I would be surprised if it were possible to do these two transformations without a visible loss in quality.
Presumably the loss at the cube face stage is somewhat less than that shown in your post? If so, the question is 'Is that quality loss more than one would expect?' I'm afraid I don't have the answer to that.
Just a thought though - which option are you selecting for 'Interpolation filter' in Pano2VR's preferences, and have you tried other settings there to seek an improvement?
Regards
Andrew
p.s. Is that Edinburgh?
360Texas, my panorama is 6000x3000 and was converted to 1910px cubes, as Pano2VR suggested by default. Thanks for explaining where this number gets from:)
Good eye Andrew, Edinburgh it is.
That's true, there were two conversions. equi->cubes->equi. Second conversion was only due to geometry - it was easier to show the difference in two equal crops. But even comparing original equirectangular and cubes version showed major quality loss. I was using Spline36 as interpolation method, tried Gaussian Sharp as well, with no luck.
When I generate flash file out of original equirectangular I get nice and sharp panorama (well, except holes in nadir and zenith). After cubes convertion it is so much worse. I know that some quality loss is a must, but isn't that too much?
Good eye Andrew, Edinburgh it is.
That's true, there were two conversions. equi->cubes->equi. Second conversion was only due to geometry - it was easier to show the difference in two equal crops. But even comparing original equirectangular and cubes version showed major quality loss. I was using Spline36 as interpolation method, tried Gaussian Sharp as well, with no luck.
When I generate flash file out of original equirectangular I get nice and sharp panorama (well, except holes in nadir and zenith). After cubes convertion it is so much worse. I know that some quality loss is a must, but isn't that too much?
- 360Texas
- Moderator
- Posts: 3684
- Joined: Sat Sep 09, 2006 6:06 pm
- Location: Fort Worth, Texas USA
- Contact:
I use SETTINGS | Images Interpolation Filter | Lanczos2 for sharpening during reassembling the cube face.
Sharp images:
Set your fisheye lens to use the hyperfocal distance value for sharpest in camera image.
http://www.dofmaster.com/dofjs.html
For a Canon 5d
Sigma 8mm at f/8
set hyperfocal focus index mark to 0.9ft or 0.03mm not at infinity oxo mark.
You can also add a tiny bit of sharpening using the on camera menu system.
In photoshop Adobe Camera Raw converter [ACR] I adjust the chromatic abberation red channel to sharpen the image.
I also adjust the white balance in ACR
In Ptgui, I use Lanczos2 sharpening algorithm to stitch the the panorama.
I suggest you try using the Lanczos sharpening algorithm when working with Pan2VR cubes.
Sharp images:
Set your fisheye lens to use the hyperfocal distance value for sharpest in camera image.
http://www.dofmaster.com/dofjs.html
For a Canon 5d
Sigma 8mm at f/8
set hyperfocal focus index mark to 0.9ft or 0.03mm not at infinity oxo mark.
You can also add a tiny bit of sharpening using the on camera menu system.
In photoshop Adobe Camera Raw converter [ACR] I adjust the chromatic abberation red channel to sharpen the image.
I also adjust the white balance in ACR
In Ptgui, I use Lanczos2 sharpening algorithm to stitch the the panorama.
I suggest you try using the Lanczos sharpening algorithm when working with Pan2VR cubes.
8mm.. I wish:) Currently I use 18mm. 3x14 multirow. I shot at f8 and focusing hyperfocal.
Sharpness of source photos or stitched panorama is not an issue, equirectangular tiff is sharp. Flash processed out from this file is also sharp. It gets blurry only after converting to cubes. And I need cubes convertion to fill nadir and zenith gaps. Thanks for the tip with lanczos2 interpolation, I will check if it makes any difference.
Sharpness of source photos or stitched panorama is not an issue, equirectangular tiff is sharp. Flash processed out from this file is also sharp. It gets blurry only after converting to cubes. And I need cubes convertion to fill nadir and zenith gaps. Thanks for the tip with lanczos2 interpolation, I will check if it makes any difference.
Nadir/Zenith
Why not run "SuperCubic" from Rune Spaans ?
http://www.superrune.com/technical/soft ... rcubic.php
It does a very good job extractin and reimporting the zenith & nadir in one go.
Also you can set your parameters to increase or decrease the FOV.
You stax in PS while retouching until the final steps are done.
edit:
For the interpolator, Jim Watters did some great tests a while ago:
http://photocreations.ca/interpolator/index.html
Why not run "SuperCubic" from Rune Spaans ?
http://www.superrune.com/technical/soft ... rcubic.php
It does a very good job extractin and reimporting the zenith & nadir in one go.
Also you can set your parameters to increase or decrease the FOV.
You stax in PS while retouching until the final steps are done.
edit:
For the interpolator, Jim Watters did some great tests a while ago:
http://photocreations.ca/interpolator/index.html