Segmentation
Mean pixel accuracy
Implemented
segmentation.mean_pixel_accuracyDefinition
Per-class pixel recall (correct pixels of the class over its true pixels), averaged over the classes that occur in the truth.
Formula
(1/K) Σ_k TP_k / (TP_k + FN_k)
Range: [0, 1]
Inputs and outputs
- y_true: ground-truth label mask
- y_pred: predicted mask with the same shape
Returns: MetricResult
Assumptions
No assumptions beyond valid, aligned inputs of the documented types.
Limitations
No metric-specific limitations are documented yet. Interpret the value alongside the task, data, and other metrics.
Python API
import evalsuite as es
es.mean_pixel_accuracy(mask_true, mask_pred)References
- Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation. CVPR 2015:3431-3440.