cellmap_flow.dashboard.routes.finetune.good_regions
===================================================

.. py:module:: cellmap_flow.dashboard.routes.finetune.good_regions

.. autoapi-nested-parse::

   Regions the user has looked at and judged the model already handles well.

   These are the counterweight to corrections. Painted scribbles say "you got
   this wrong, here"; a good region says "you got this right, here, and I
   checked". Training samples patches from both: the scribbles supply the
   supervised signal, and the good regions supply rehearsal -- patches with no
   annotation at all, which the trainer already treats correctly (an empty mask
   zeroes the supervised term and hands the whole patch to the distillation
   term against the frozen teacher).

   Kept deliberately separate from ``imported_crops``: those carry real labels,
   these carry only a human's assertion that the model's own output is
   acceptable there. Both are evidence, but not the same kind.

   Routes: GET ``/api/finetune/good-regions`` (the session's regions), POST
   ``/api/finetune/good-regions/mark-view`` (a box where the viewer looks) and
   POST ``/api/finetune/good-regions/delete`` (one region by ``id``, or all).



Attributes
----------

.. autoapisummary::

   cellmap_flow.dashboard.routes.finetune.good_regions.logger
   cellmap_flow.dashboard.routes.finetune.good_regions.GOOD_REGIONS_LAYER
   cellmap_flow.dashboard.routes.finetune.good_regions.DEFAULT_REGION_SIZE_NM


Functions
---------

.. autoapisummary::

   cellmap_flow.dashboard.routes.finetune.good_regions.load_good_regions
   cellmap_flow.dashboard.routes.finetune.good_regions.save_good_regions
   cellmap_flow.dashboard.routes.finetune.good_regions.view_box_nm
   cellmap_flow.dashboard.routes.finetune.good_regions.mark_current_view_good
   cellmap_flow.dashboard.routes.finetune.good_regions.list_good_regions
   cellmap_flow.dashboard.routes.finetune.good_regions.delete_good_region
   cellmap_flow.dashboard.routes.finetune.good_regions.refresh_good_regions_layer


Module Contents
---------------

.. py:data:: logger

.. py:data:: GOOD_REGIONS_LAYER
   :value: 'good_regions'


.. py:data:: DEFAULT_REGION_SIZE_NM
   :value: 896.0


.. py:function:: load_good_regions()

.. py:function:: save_good_regions(regions)

.. py:function:: view_box_nm(size_nm=None)

   ``(centre_nm, size_nm)`` of the box centred where the viewer looks.

   ``size_nm`` defaults to one model output patch (see
   DEFAULT_REGION_SIZE_NM); one number is the same along every axis.
   The labelling buttons beside "Mark This View as Good" use the same box,
   so what they label is what a mark would cover. ValueError when there
   is no viewer position or the size is not positive.


.. py:function:: mark_current_view_good()

   Record a box centred on wherever the viewer is looking right now.


.. py:function:: list_good_regions()

.. py:function:: delete_good_region()

.. py:function:: refresh_good_regions_layer(regions=None)

   Draw the good regions in the viewer so you can see what you marked.