Self-Calibration differs from Field Calibration in that field calibration works with a single camera in the project and the result is a calibration of a single camera, whereas self-calibration can start with one or more cameras in the project and results with a special ‘self-calibration’ camera per photo. This is used when the camera parameters might be different for every photo. Note that self-calibration adds many more parameters to the solution (a set of camera internal parameters per photo) and works best if there is a lot of input data (many referenced marked points on the photos). This also means that generally self-calibration works with project with a reasonable number of photos (perhaps under 15 photos).
This switch between field and self-calibration depends on the number of cameras in the project and the Preference settings (Preferences - Processing and Cameras - “Calibration in one camera projects is a full field calibration” setting).
When the camera optimization is self-calibration vs field-calibration, the Processing Dialog will show the corresponding text in the Camera Optimization section.
When one camera is shared between multiple photographs, self-calibration copies the camera (one for each shared photo) during processing so they can be independently adjusted. After processing is completed, you can see the results of the individual calibrations by opening the Photo Table (the camera names in the Camera column will have :scN appended – and you can see the variations in the camera parameters in their respective columns). These individualized camera settings will remain until the next processing or until a camera parameter is changed on the original camera for that photograph.
Note that because self-calibration has many parameters to solve it is inherently less stable. To help reduce the range of camera parameter solutions and to stabilize the solution, there is a second setting in the Preferences called “Self-calibration uses Std. Dev. Weights”. When this is checked the values below in the Preferences are used to weight the camera parameters (the smaller the value, the greater the weight and the less that parameter can change).