Forum: Tips & Tricks

Thread: optimizing CPDs

back to threads list | this thread is closed
Search the forums:
Author  Message 

Posts: 336
Location: Antalya, Turkey
Registered: 7 Jun 2012
optimizing CPDs
Posted: 14 Oct 2012 at 18:42 GMT
updated: 14 Oct 2012 at 18:47 GMT
Throu coincidence I have made the following discovery:
I have my 4 shots around...Put them into PTGUI, align the images and optimize...I get an aver. CPD...It is mostly in the area of "very good"...To make this result better I use the function "delete the worst CPs"...PTGUI often doesn't find any statistical outliners and doesn't delete any or just a few CPs...Then I create new CPs by using "generate new CPs"...I optimize...Then I delete the worst CPs...This time PTGUI finds CPs to delete...And the resulting aver. CPD of the optimizing process is better then before generating new CPs...The lucky coincidence hit me about what I do next...In the advanced mode of the optimizer I uncheck one of the images (nr.4)...Without optimising again I click on "delete the worst CPs"...PTGUI evaluates only the three images and finds often many CPs to delete...And the resulting aver.CPD is much better...Then I check the box next to Nr. 4; I take it into the game again...As next step I uncheck the image NR.3...Delete worst CPs...Do the same for all the images ...This is now my workflow and I get CPDS I never saw before...If one takes time and repeats the whole process there's chance for even better CPDS...I just got from 1,70 in the beginning to 0,54 aver. CPD...
alert moderator
John Houghton

Posts: 3934
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
Re: optimizing CPDs
Posted: 14 Oct 2012 at 19:03 GMT
It's easy to get a low average control points distance by deleting points, but that doesn't necessarily yield a perfect stitch. You also need to maintain a good spread of points all along the overlap areas. If there are no points in the corners of the images, then the images may well not align properly in those areas.

alert moderator

Posts: 4310
Location: Mount Panorama, Australia
Registered: 21 Jul 2004
Re: optimizing CPDs
Posted: 15 Oct 2012 at 5:19 GMT
You need many control points with a low CPD. These control points must be from top to bottom of the image overlap as John stated. Clustered controls points are of little use.

You should also be able to manually place control points and still yield a low CPD.

Your keyboard arrows allow you to nudge selected control point by the pixel. Keyboard F5 to re-optimize.

Holding down the keyboard keys Ctrl and Alt together will help you find exact position placement using cross hair and zoom.

It is good to know you are experimenting, learning and finding your way with PTGui.

Regards, Smooth cool
alert moderator

Posts: 336
Location: Antalya, Turkey
Registered: 7 Jun 2012
Re: optimizing CPDs
Posted: 15 Oct 2012 at 10:17 GMT
updated: 15 Oct 2012 at 10:18 GMT
And I was wondering why I get bad stitching results in a project although brilliant aver. CPD smile...I looked in the project again and where the stitching was bad, there were no CPs at all...Adding there selectively on that region automatically CPs didn'T work PTGUI couldn't find any...So I placed them manually...And the result is much much better...But I think PTGUI doesn't like the new CPssmile...So I don't use the option delete the worst CPs after optimizing...He first gets rid of the new CPs - at least of many of themsmile) Besides I see now an overall good distribution of CPs is also a must for a good stitching resultn - it is not about only "the problem area"......John, Smooth, thank you very much for taking time...
alert moderator

Posts: 1787
Location: Los Anglels, United States
Registered: 1 Sep 2007
Re: optimizing CPDs
Posted: 15 Oct 2012 at 13:47 GMT

You may be overthinking the control point distance.

As long as you have a good spread along each seam and the panorama stitches without error, you can pretty much ignore the control point distance statistics.

Take a look at This pano has a maximum control point distance of 16. The average is 2.9 and the minimum is 0.1. I could probably get the one or two points that are way out to have a lower number, but that would not improve the stitch. If the pano did not stitch, the numbers are irrelevant since my rig would need better calibration. You use the numbers as a guide. You use your eyes to make the determination if your pano stitched or not. The numbers are important, but not the final word.

In the Panorama Editor window, turn on the seams and use the detail viewer to examine each seam. If everything is aligned and your pano is level, you are done. Getting a lower maximum control point distance number is nice, but not a requirement for a good stitch.

Sometimes you have features in your pano that cause PTGui to assign control points to images that are not next to each other. Look for and remove control points in images that are not next to each other. The Eiffel tower is an example of a structure that can be a nightmare to stitch due to the repeating structural features.

I am going to spend time dialing in my rig to see if I can get the numbers lower, but I am not going to obsess over it. Monopod shooting is inherently inaccurate. All the panos is the series stitched just fine. I excluded over half of what I shot not because of stitching errors, but because of boring content.

alert moderator