`LibMTL.metrics`¶

class AbsMetric[source]¶

Bases: object

An abstract class for the performance metrics of a task.

record¶

A list of the metric scores in every iteration.

Type:: list

bs¶

A list of the number of data in every iteration.

Type:: list

property update_fun[source]¶

Calculate the metric scores in every iteration and update :attr:`record`.

Parameters:

pred (torch.Tensor) – The prediction tensor.
gt (torch.Tensor) – The ground-truth tensor.

property score_fun[source]¶

Calculate the final score (when an epoch ends).

Returns:: A list of metric scores.
Return type:: list

reinit()[source]¶: Reset record and bs (when an epoch ends).

class AccMetric[source]¶

Bases: AbsMetric

Calculate the accuracy.

update_fun(pred, gt)[source]¶

score_fun()[source]¶

class L1Metric[source]¶

Bases: AbsMetric

Calculate the Mean Absolute Error (MAE).

update_fun(pred, gt)[source]¶

score_fun()[source]¶