cellmap_flow.ai_annotate.backends.fake

A provider that runs locally and calls nothing, for tests and demos.

It stands in for a hosted model wherever the dashboard’s own behaviour is what is being checked: the review panel, Accept and Reject, the undo. Its mask is deterministic and plausible on EM: the pixels darker than the plane’s median within a disk around the click, which on most EM picks out membranes and dense organelles near where the user pointed.

Classes

FakeBackend

Module Contents

class cellmap_flow.ai_annotate.backends.fake.FakeBackend

segment(request: cellmap_flow.ai_annotate.backends.base.SegmentRequest, model: str) → cellmap_flow.ai_annotate.backends.base.SegmentResult

Dark pixels within min(H, W) / 4 of the click (or the centre), and the input with them painted target_rgb as the “model’s” image.