Controlling control points.

[Velson Horie]

I am stitching a 2 by 8 mosaic, see screenshots

After I loaded the tiles, I loaded a 2x7 template (it seems that loading a template before the images causes problems), then changed it to a 2x8 template (which I saved).

I ticked the "Images are already roughly  ;.."

then "Align images", then "Generate control Points"

The first control panel screenshot shows the output after choosing to create control points between adjacent tiles 1 and 2. No CPs created.

CPs for the adjacent tiles 1 and 16 were created by Create CPs, but are unrepresentative so unlikely to stitch with uniform error across the overlap.

It would seem sensible to specify that CPs are created in a roughly uniform distribution, e.g. grid, across the overlap, defined by PTGui or me. I frequently find myself putting CPs in areas empty of CPs, where there are significantly mismatched stitching lines.

In the 1+2 tile example, it would be useful to force the creation of CPs. It is obvious to my eye and presumably the feature matching tool of the program that there are features common to both.

Velson

[Erik Krause]

Am 18.09.2024 um 16:08 schrieb Velson Horie:

> I am stitching a 2 by 8 mosaic, see screenshots

What focal length did you specify? In your screenshots it looks like
scanned images, which are in orthographic projection. To mimic that in
PTGui you would need to specify a very long focal length. See
https://ptgui.com/support.html#6_5 for details.

It might be, PTGui assumes a relatively short focal length, and hence
tries to compensate for perspective distortion, which would lead to
misalignments. Control points are restricted to areas with little
misalignment, hence probably the bad distribution.

Only wild guesses, though. For a better analysis we'd need your project
file and eventually smaller copies (half size jpegs) of your source images.

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

[Velson Horie]

Thank you.

It looks as though the problem was the lens settings. I thought that I had reset the lens corections in the 2x7 grid template I had previously created, then input into this project, then input the images, then changed from a 2x7 grid to a 2x8 grid. But I did not recheck these settings or indeed the Optimiser parameters.  Both of these appear to have been changed somewhere along the line.

What is the correct procedure for saving and importing a template then modifying it? 

As an aside:

I want to check that the CPs with large errors reported are incorrect.

In the screen shot is a CP with a 20.2 distance. I want to flick back and forth between the CP panel and the Detailed view panel to inspect the detailed position in its context.  So it would be really useful to move the cursor between the panels without moving the mouse.  Moving the mouse loses the position in the Detail Viewer.

Actually finding the position on the Detail viewer is a pain. Clicking on the (useful) CP table takes the cursor to the CP shown on the pair of tiles.  But then locating the CP position on the Detail viewer requires some semi-blind navigation.

[PTGui Support]

On 19-09-2024 16:05, Velson Horie wrote:
> What is the correct procedure for saving and importing a template then
> modifying it?

The template can be opened and saved in PTGui as any regular project
file (File - Open Project, File - Save Project).

> As an aside:
> I want to check that the CPs with large errors reported are incorrect.
> In the screen shot is a CP with a 20.2 distance. I want to flick back
> and forth between the CP panel and the Detailed view panel to inspect
> the detailed position in its context.  So it would be really useful to
> move the cursor between the panels without moving the mouse.  Moving the
> mouse loses the position in the Detail Viewer.
> Actually finding the position on the Detail viewer is a pain. Clicking
> on the (useful) CP table takes the cursor to the CP shown on the pair of
> tiles.  But then locating the CP position on the Detail viewer requires
> some semi-blind navigation.

If you zoom out the detail viewer (so the entire image is visible), and
hover the mouse over the CP tab, you will see a 'ghost mouse cursor' in
the detail viewer. The ghost cursor is synchronized to the mouse
position in the CP tab, this would show you where to zoom in in the
detail viewer.

Joost

[Velson Horie]

Joost

Thank you

Templates

I have constructed a number of general templates, depending on the number of tiles to be stitched e.g. 1x2template, or 2x8template.  In those, I have set up the recommended settings , reset the lens settings, optimise clear all global lens profice, and optimise each image except one, etc.  These are saved in the default folder C:\Users\...\AppData\Roaming\PTGui\Templates

I have just examined the text of my 2x7template, and find that it specifies the tile file names for latest panorama using that template.  I have not completely teased out what happens when I start a new stitching project. If I load the template then load a new set of tile files, PTGui seems to apply different settings than if I load the tile files then the template.  Or it may be that my processes are not consistent.

Shouldn't a template be independent of any specific instance?

The specific instance is described in the Project File.

CP and Detail View

I agree, and the ghost mouse cursor can (just) be seen in the screen shot I sent.  The finished panorama was 47 GB.

1.  Given the complex random features in the CP panel and DV, low distance proposed CPs in the two tiles are often not the same point on the object or even near (this is a common occurance, which is why each CP needs to be checked individually).  It is not possible to move the mouse to either of the CP positions of the two tiles in DV. My visual memory is not precise enough to place the mouse identically in this noisy environment. 

2. To improve the matching of CP placements in tiles, I usually work at 100% view.  I click on a CP listed in the CP table to assess the match on the tiles.  To navigate to that point in DV, I have to zoom out DV from 100% to about 4%, by moving the mouse away from the tiles and using the wheel to change magnification, placing the mouse back on the CP panels to relocate the ghost cursor in the DV,  then focus down again, scanning across the next magnification of DV to find the cursor, then repeat till DV is back at 100% and the ghost cursor on one of the CP options.  

When I click on the next CP in the table, the ghost cursor (positioned at the CP panel position) knows where it is on the DV (somewhere in 2 tiles across a field of 40,000x30,000 pixels - and I don't).  Would it be possible to just click something (probably keyboard) to move the mouse to the ghost cursor position on the DV at the CP magnification, while maintaining the matching cursor position on the tiles showing the CPs? And then to do the reverse?

I go through a similar process, but starting from the DV, when improving mis-alignment of a stitching line by placing a CP near the line.  Navigating from a line's position on DV to the matching one(s) on the tiles at 100% is not straight forward.

Velson

[PTGui Support]

Hi Velson,

On 20-09-2024 13:04, Velson Horie wrote:
> I have just examined the text of my 2x7template, and find that it
> specifies the tile file names for latest panorama using that template.

That's correct; this is because the template is a regular project file.
When applying the template, the file names are ignored.

> Navigating from a line's position on DV to the matching one(s) on the
> tiles at 100% is not straight forward.

Agreed. And it would be useful to have a feature for navigating between
a specified point in one window to the same point in other windows. I'll
add this to the wish list.

Joost

[Velson Horie]

Joost

Thank you.  I imagine that the users' combined wish list is extensive and diverse (even contradictory!).

PTGui's matching of features then placing CPs is, to me, extraordinarily accurate. I therefore want to game the method to make it more effective and hopefully more efficient for my application.

The first screenshot shows that CP1 is in the correct position, probably to within 2-3 pixels.  Yet the CP table opines that it is 26 pixels out.

It also shows that the cursor position on the two tiles is out by ca 50 pixels.

The second screenshot shows that the new CP9 is nearly in the correct position, probably to within 6-8 pixels (and I could adjust the position to improve the accuracy). Yet the CP table opines that it is 29 pixels out.  Presumably the tool has decided on a mid-point between the two observed cusor positions.

PTGui must have constructed a base world coordinate system that is about 27 pixels out, at least in this area.

If I could tell PTGui that CP1 and 9 are correct within +/- 2 pixel, could that be used to modify the base world coordinates in this area?

Which brings me back to my previous suggestion. Can we tell PTGui to produce CPs on a grid, which after confirmation would then be used to create local coordinate systems that would flow into each other?  External geometric fixings such as horizons, straight lines etc would set the limits of this flow.

Velson

[PTGui Support]

On 21-09-2024 16:31, Velson Horie wrote:
> PTGui must have constructed a base world coordinate system that is about
> 27 pixels out, at least in this area.

Exactly. PTGui optimizes the yaw/pitch/roll and lens settings such that
the errors are minimized. If the images do not fit this model, errors
will remain.

> If I could tell PTGui that CP1 and 9 are correct within +/- 2 pixel,
> could that be used to modify the base world coordinates in this area?
>
> Which brings me back to my previous suggestion. Can we tell PTGui to
> produce CPs on a grid, which after confirmation would then be used to
> create local coordinate systems that would flow into each other?
> External geometric fixings such as horizons, straight lines etc would
> set the limits of this flow.

No, currently PTGui cannot do this. It might be interesting to explore
this in a future update though.

Joost