Troubleshooting Papywizard Cases
Papywizard module is really robust, but in some cases, issues can happen and it is important to know how to address them.
Here, we will publish examples of a problematic papywizard cases and how to quickly fix them.
Case 1 - Exif issue and dust
First, here's a screenshot of a direct detection without any attention (every settings were in default mode). I did highlight in red some of the visual glishes.
To address all issues, the best is to go in a more manual mode (meaning, we'll disable automatic algorithms to inspect indivudal issue at each stages).
In Papywizard plugin
First, have a look to the warning that Papywizard already tells you. This first warning is quite critical because it says that the focal length found in EXIF is different that the one found in XML file. This case happen because in some cases : lenses without electronic connection to the camera, wrong setting in papywizard software, etc.
Here, you need to say to the plugin which focal length is the real one. Remember that focal length in Autopano is quite always expressed as 35mm focal equivalent (you need to take care of the sensor scale to get this value).
To do this, hit the "Shoot setup" tab and use the right button "Use XML focal length" or "Use Image focal Length" : In this case, it was the XML that was right.
To be able to analyse control point detected, we'll switch off the optimizer, so that in the editor, input images will be located at the recorded location and not moved according to the control point (optimization stage is skipped). To do that, use th "Detection options" tab to check "skip optimization".
In the editor
Open the editor on the detected panorama and activate the control point editor. You'll get such a view with blue links.
How to read such a view ?
With the papywizard plugin, we'll inspect all images to see if they are overlapping or not (this is done using 2 inputs values : the xml coordinates that tells where each images are located, and the focal length, which says how large each image are). If 2 images are overlapping, then, we'll try to find control point between them. If not, the link will not be calculated. It happens that 2 images are still overlapping but no control point are found (typically in the sky). In this case, we still add a single control point so that the sky will follow the ground if the ground deforms itself because of real control point. That's why you can see a lot of links with just "1" cp. These are hard links so that the sky stay ok when optimizing the ground.
As something went wrong, it is often because of wrong links. We need to check them. Wrong links can come because of some highly symmetric object that can confuse the detector, because of lack of details, etc. As we used XML, each real link should have a quite low RMS already because we are not far from real solution. A typical value is around 20. So any link that has a value far beyond 20 is suspect.
How to simplify the view to view only suspicious links ?
Just use this slider in the editor. Set the value to 20. Then, you'll get the following editor view :
You have some really rms links, with value quite high. Let's inspect them (Note : ignore the NaN cases, there are not relevant here).
Let's just click on one of these wrong links:
You can see some strange control points. Why are they there ? Because of dust. In this case, a lot of dust were remaining on the lenses. As autopano didn't have any relevant details, it found dust instead of real content like on the ground. So we need to remove these wrong links. You can do it manually, by selecting the links manually one by one by clicking on the shaure between images or you can use a more automatic way.
The automatic way is easy. You have button below the slider that hides links which will delete all links that are still visible. As you set the value to 20, hitting that button will erase any links over that value. Let's do that.
As you changed control points, you need to say to autopano to take into account the new control points. To do that, you need to optimize the panorama. Let's do that.
It is already quite okay, but there is still one issue : the top part of the panorama is not linked to the bottom part of the panorama. It seems that one issue happened there but we don't know the real cause.
Nevertheless, it is quite easy to solve. In control point editor, right click on one image you want to be linked to another one. You'll get the following menu :
What is interesting for us is the first entry. The local geometrical analysis will inspect all images overlapping the one you clicked on and will check if it can find some control point.
Do that with several images, you don't need all of them, just some regularly spaced in the panorama. You should end up with the following view :
Last step is to optimize again the panorama to get the final geometric aspect for this case.
The global RMS is 1.49, which is an absolutely perfect value. Final rendering is the following :
The total time to handle this case is around 5 minutes from initial detection to final perfect project.