This is a tool for pixel-precise, semi-automated image annotation.
To speed up usage, please consult the keyboard shortcuts.

There are 3 kinds of input for this tool: images, masks and preannotations.
Images are always required: to load an image, select an image folder in the
leftmost list and then double click on any entry. The corresponding image
should appear in the right area. You can then navigate and zoom using the
mouse wheel and right button.

With each image, an empty "mask" and "preannotation mask" is loaded. You
can load a previously saved mask by selecting an image in the second
list. Masks are images where very non-zero pixel is considered a defect.

Preannotations are matrices of floats, usually telling the probability of
being a defect for each pixel. By using the "Thresh" slider, the user can
decide which elements will show up as defects in the preannotation mask.
The magnitude of the slider is in acceptance p-value, i.e., a value of
p means that the top p region of the preannotation values will be considered
as defects. Load the preannotations with the left list.

Currently the following tools are supported to edit the masks:

* Painter: Paints a circle of specified diameter into the specified mask
* Eraser: Opposite function to painter
* Masked painter: Like painter, but paints only in the positions where the
  preannotation is active.

The tool provides undo/redo functionality to all of them.

If the preannotations have many false positives, the masked painter is most
efficient. Otherwise erasing the preannotations is also an option.
Once you are done, select the output folder at the bottom region, as well as
the other options, and save the masks. They can be loaded into the second
list in future sessions, or directly used in further applications.

The masks colors are there only for convenience. They can be changed to help
visualize the defects.

This tool is a prototype. Please see the about section for any bug reports or
inquiries. Thank you!
