Modify warping tile priority

[ganctom]

There is inherent and frequent issue of distorted few first lines of every EM image (below top edge). So far, the warping algorithms, have been warping each stitched vertical tile-pair with bottom image overlaid over its top neighbor. In order to mask the distortion, a reversed warping priority should cover most of these distortion with image information from the top neighbor.

[tibuch]

How many rows are affected by this and to what value did you set the margin parameter? And how much is the tile-overlap?

[ganctom]

The distortion can cover up to 50 lines, depending on factors such as sample conductivity, electron dose, and cutting thickness. This phenomenon is not deterministic but significantly impacts data quality. Too much charging can lead to even larger distorted portions of the image.

Each section has an upper limit for the margin parameter. Depending on computed xy shifts, a too large margin will cause visible gaps (missing information) during tile stitching. Currently, margin is a scalar constant for the entire section.

Ideally, for each tile-pair, the margin should be as large as possible while still preventing gaps. The allowed margin value directly depends on the size of the shift vector (its vertical or horizontal components, to be more precise).

The tile-overlap is usually set to 200 pixels, but in reality, it can be as small as a few tens of pixels for complex sections.

[tibuch]

There is an option to provide margins for left, right, top and bottom of a tile. This parameter can be exposed.

I could not find an option to define which tile goes on top or bottom.

[ganctom]

Please, expose that parameter for testing purposes. Thank you.