segmetrics.study

class segmetrics.study.Study

Bases: object

Computes different performance measures for different image data.

Performance measures must be added prior to performing evaluation.

add_measure(measure: MeasureProtocol, name: str | None = None) → str

Adds a performance measure to this study.

Parameters:

  • measure – The performance measure to be added.

  • name – An arbitrary name which uniquely identifies the performance measure within this study. Uses measure.default_name() if None is given.

Returns:

The name used for the measure (see above).

merge(other: Study, sample_ids: Sequence[Any] | Literal['all'], replace: bool = True) → None

Merges measures and results from other study.

Parameters:

  • other – The study which is to be merged into this study.

  • sample_ids – The identifiers of the samples which are to be merged (or 'all').

  • replace – Whether conflicting identifiers are to be replaced (True) or prohibited (False).

print_results(write: ~typing.Callable[[str], ~typing.Any] = <built-in method write of _io.TextIOWrapper object>, pad: int = 0, fmt_unbound_float: str = 'g', line_suffix: str = '\n') → None

Prints the results of this study.

Parameters:

  • write – Function used for output.

  • pad – Number of whitespaces at the start of each line.

  • fmt_unbound_float – Formatting literal used for decimals not presented as percentages.

  • line_suffix – Suffix of each line.

process(sample_id: Any, actual: None, unique: bool = True, replace: bool = True) → Dict[str, List[float]]

Evaluates a segmentation result based on the previously set expected result.

If unique is True, it is assumed that all objects are labeled uniquely. Use unique=False if this is not guaranteed and the individual objects should be determined by connected component analysis instead (e.g., if actual is a binary image which represents the union of the individual object masks).

The image actual must be a numpy array of integral data type. It is also allowed to be boolean if and only if unique=False is used.

Parameters:

  • sample_id – An arbitrary indentifier of the segmentation image (e.g., the filename).

  • actual – An image containing object masks corresponding to the segmentation result.

  • unique – Whether the individual object masks are uniquely labeled. Providing False assumes that connected components correspond to individual objects (components of different labels are not connected).

  • replace – Whether previous results computed for the same sample_id should be replaced (True) or forbidden (False).

reset() → None

Resets all results computed so far in this study.

set_expected(expected: None, unique: bool = True) → None

Sets the expected ground truth segmentation result.

The background of the image must be labeled as 0. Negative object labels are forbidden. If unique is True, it is assumed that all objects are labeled uniquely. Use unique=False if this is not guaranteed and the individual objects should be determined by connected component analysis instead (e.g., if expected is a binary image which represents the union of the individual object masks).

The image expected must be a numpy array of integral data type. It is also allowed to be boolean if and only if unique=False is used.

Parameters:

  • expected – An image containing object masks corresponding to the ground truth.

  • unique – Whether the individual object masks are uniquely labeled. Providing False assumes that connected components correspond to individual objects (components of different labels are not connected).

todf() → pd.DataFrame

Returns the results of this study as a pandas dataframe.

write_csv(fout: TextIO, write_samples: bool | str = 'auto', write_header: bool = True, write_summary: bool = True, **kwargs) → None

Writes the results of this study as CSV.

Parameters:

  • fout – File descriptor used for output.

  • write_samples – Whether all samples should be written separately.

  • write_header – Whether a header should be included (measure names).

  • write_summary – Whether a summary should be included in the last row.

  • kwargs – Additional parameters for csv.writer, see: https://docs.python.org/3/library/csv.html#csv.writer

write_tsv(fout: TextIO, **kwargs) → None

Writes the results of this study as TSV.

Parameters:

  • fout – File descriptor used for output.

  • kwargs – Additional parameters passed to write_csv().