Backend-agnostic prediction facade. The model's backend selects the
protocol (stardist.2d.v1 or cellpose.2d.v1); the run goes through
sg_protocol_run() so parameters are validated and recorded, the
backend is checked at the point of use, and the resulting mask is
staged with the model digest in its provenance. Predictions never
overwrite reviewed masks; use sg_replace_mask() to adopt a reviewed
result explicitly.
Usage
sg_predict_model(model, image, ..., protocol = NULL, output = c("mask", "run"))Arguments
- model
An
sg_trained_modelor the path of a.segmantRarchive (loaded with integrity verification).- image
An
sg_image.- ...
Protocol parameters (see
sg_protocol_schema(protocol)).- protocol
Optional protocol id overriding the backend default.
- output
"mask"or"run"(seesg_protocol_run()).
Examples
# \donttest{
mdl <- new_sg_trained_model(tempfile("model"), backend = "stardist",
base_model = "none", training_metrics = list())
img <- sg_example_image("fluorescence_nuclei")
run <- sg_predict_model(mdl, img, output = "run")
# "unavailable" unless a Python with 'stardist' has been selected
# (see sg_setup_python())
run$record$status
#> [1] "unavailable"
# }