Image quality issues.

Q&A about the latest versions
Post Reply
DemonDuck
Posts: 111
Joined: Sun Feb 15, 2009 1:15 am

I have an 8800x4400 equirectangular panorama. I set the screen size to 1600x1600. I set the cubeface size to optimal -- 2285. I set the Render Quality to 12 -- the highest.

When I look at the equirectangular image in FSPViewer and compare to the output from Pano2VR -- FSPViewer has slightly better image quality at approximately the same zoom level.

What can I do and what settings should I use to get the same image quality as FSPViewer?
User avatar
360Texas
Moderator
Posts: 3684
Joined: Sat Sep 09, 2006 6:06 pm
Location: Fort Worth, Texas USA
Contact:

Your image is 8800 wide/ Pi (3.1415)= 2801.2096

You might try setting the cube face size to 2801px to increase the image quality.

Of course... you are not showing us the two images.. so really can not tell for sure.

"Slightly better image quality" is a judgemental term.

Of course you probably already knew this... and were just teasing us.

Still lost in Mexico
Dave
Pano2VR Forum Global Moderator
Image
Visit 360texas.com
DemonDuck
Posts: 111
Joined: Sun Feb 15, 2009 1:15 am

360Texas wrote:Your image is 8800 wide/ Pi (3.1415)= 2801.2096

You might try setting the cube face size to 2801px to increase the image quality.

Of course... you are not showing us the two images.. so really can not tell for sure.

"Slightly better image quality" is a judgemental term.

Of course you probably already knew this... and were just teasing us.

Still lost in Mexico
I set the window size to 2000x2000 that gives an optimal cube face size of 2856. Using that I made a test using image quality 100% and sub tiles 1.

A screen capture comparing FSPViewer with Pano2VR shows only a slight degradation of sharpness and a slight color shift. Pano2VR is bluer.

The differences are so slight that most of you will not be able to see the difference and you will become harsh and say I'm crazy or too fussy -- so be it.

http://pancyl.com/images/ScreenCapture.tif
erik leeman
Posts: 470
Joined: Sat Sep 09, 2006 10:51 am
Contact:

Don't worry, the differences are quite obvious!
You could try a different Interpolation Filter in Pano2VR by changing it in the Application Settings menu.
The Mammoth Rock on the right looks oversharpened to me, but if that's what you like you could use Lanczos 2 or 3.
What Interpolation Filter are you using now in both applications?

One other thing, if your equirect is (only) 8800x4400 pixels in size, you shouldn't make your cube faces larger than 2800x2800.
If you do make them larger even the pixels in the center of those cube faces will have to be 'stretched', and image quality will suffer.
In stead, make them a bit smaller than 2800x2800, and things will probably look better already.

Erik
DemonDuck
Posts: 111
Joined: Sun Feb 15, 2009 1:15 am

erik leeman wrote:Don't worry, the differences are quite obvious!
You could try a different Interpolation Filter in Pano2VR by changing it in the Application Settings menu.
The Mammoth Rock on the right looks oversharpened to me, but if that's what you like you could use Lanczos 2 or 3.
What Interpolation Filter are you using now in both applications?

One other thing, if your equirect is (only) 8800x4400 pixels in size, you shouldn't make your cube faces larger than 2800x2800.
If you do make them larger even the pixels in the center of those cube faces will have to be 'stretched', and image quality will suffer.
In stead, make them a bit smaller than 2800x2800, and things will probably look better already.

Erik
In PTGui I use Lanczos2. In the first screen shot of Pano2VR, the setting was for Blackman/sinc. In this second screen shot I changed the setting to Lanczos2 in Pano2VR. It appears to be even less sharp and have more of a color shift.

http://pancyl.com/images/ScreenCapture2.tif

I also made the cube faces 2801x2801. Overall, the first version seems better.

Let me suggest this hypothesis. The source file is jpeg because I used the free version of NeatImage to de-noise and the free version only saves in jpeg. So when Pano2VR converts the equi to cube faces, that's two jpeg saves. That would degrade the image. Yes? No? Comments?

And yes, I do tend to over sharpen. I try to take sharpening right to the limit of sensibility. Sometimes I go to far. I only sharpened in NeatImage and really didn't use to much sharpening. But looking really close (the screen shot is 131%) I agree -- a little too much sharpening.

Here's the whole image. Who can make the best image quality Flash presentation?
http://pancyl.com/images/MammothRock.jpg
User avatar
Hopki
Gnome
Posts: 13021
Joined: Thu Jan 10, 2008 3:16 pm
Location: Layer de la Haye, Essex UK
Contact:

http://support.ggno.me/dduck

Pano2VR V3.1.0
Interpolation Filter: Blackman/sinc
Flash Output: Cube Face Size 2800
Image Quality 100%

Multi Res Settings:
Level Tile size: 175px

Levels:
2800
1400
700
350
175
No Embed, No Load at start up and No Decode at start up.

SWF size 78.3KB
Tiles folder size 77.7 MB

Not very practical if you have bandwidth restrictions.
Hopki
Garden Gnome Support
If you send an e-mail to support please send a link to the forum post for reference.
support@ggnome.com
https://ggnome.com/wiki/documentation/
User avatar
360Texas
Moderator
Posts: 3684
Joined: Sat Sep 09, 2006 6:06 pm
Location: Fort Worth, Texas USA
Contact:

There are strong reasons for NOT using JPG's for creating panoramas.

RAW of course is the natural selection for original image sets

Our source original for Pano2vr is
6000 x 3000,
16bit,
white balanced,
chromatic abberation fixed (Sigma 8mm fisheye),
uncompressed TIF
Produces a 100 + mb equirectangle

I think I was told once that when loading a source 16bit image, Pano2vr as a FINAL step converts the finished image set to 8bit anyway.

JPEG source images probably have artifacts because of multiple Saves and Sharpening:

Saving:
in camera jpg save compression,
stitched saving and re-compression
post processing saving and re-compression

Sharpening:
in camera during Jpg save to memory (unless you have in-camera sharpening turned off)
during stitching using the stitchers algorithm interpolator
post processing sharpening like PhotoShop's native interpolator
Pano2vr processing user selected interpolator

It is recommended that 8bit jpg's as source images should be avoided when you are expecting 'super' sharp panoramas

Again, you probably already know all this because you have been creating panoramas for years. You created a panorama viewer several years ago. I have seen your panorama viewer at work.. it rotates smoothly and renders a very sharp image. It makes between pano transitions very smoothly. I think it operates in the java environment. For others that have not yet visited your site: http://pancyl.com/ Here you can see a similar panorama ... only this one is a very well created cylinder.

Your PanCylApplet.jar javascript code set is very similar to Helmet Dersch's PTviewer java library Class early work.
Dave
Pano2VR Forum Global Moderator
Image
Visit 360texas.com
DemonDuck
Posts: 111
Joined: Sun Feb 15, 2009 1:15 am

360Texas wrote: Again, you probably already know all this because you have been creating panoramas for years. You created a panorama viewer several years ago. I have seen your panorama viewer at work.. it rotates smoothly and renders a very sharp image. It makes between pano transitions very smoothly. I think it operates in the java environment. For others that have not yet visited your site: http://pancyl.com/ Here you can see a similar panorama ... only this one is a very well created cylinder.

Your PanCylApplet.jar javascript code set is very similar to Helmet Dersch's PTviewer java library Class early work.
Yes, and I agree that jpegs are not the best source. This is a unique situation because I'm using a trial version of NeatImage (which is very good) and the trial version saves only in jpeg.

Thanks for mentioning my cylindrical viewer. But it's a dead viewer and a dead site. It's java -- who cares? But it was the only pure cylindrical viewer that I know of using a pure cylindrical projection -- not a clipped spherical projection with tilt compensation

http://pancyl.com/MarinaCity.htm

Zoom in a little so you can tilt up and down.

I also wrote a spherical viewer -- PanGnomic -- which uses a pure gnomic projection. And it's a pixel mapper not a texture mapper. I think that's different than what Flash viewers use today.

But that's all history. You can compare viewers -- Java to Flash here:
http://www.panoramaphotographer.com/comparisons/web/
DemonDuck
Posts: 111
Joined: Sun Feb 15, 2009 1:15 am

Hopki wrote:http://support.ggno.me/dduck

Pano2VR V3.1.0
Interpolation Filter: Blackman/sinc
Flash Output: Cube Face Size 2800
Image Quality 100%

Multi Res Settings:
Level Tile size: 175px

Levels:
2800
1400
700
350
175
No Embed, No Load at start up and No Decode at start up.

SWF size 78.3KB
Tiles folder size 77.7 MB

Not very practical if you have bandwidth restrictions.
Hopki
Yeah, the size of the presentation is a killer. I actually got better results using just a single tile of 2865x2865 which results in a .swf of 48.415 -- to me that's preferable but that's about a 4 times bloat of the image which is 11,480. I'm probably not packaging it right.
Post Reply