[Validate] Fix segmentation confusion masking (#330)

* Copy rows/cols from confusion when filtering instead of masking

* Bump and changelog

Author: gatli

CHANGELOG.md

@@ -5,6 +5,11 @@ All notable changes to the [Nucleus Python Client](https://github.com/scaleapi/n
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.14.8](https://github.com/scaleapi/nucleus-python-client/releases/tag/v0.14.8) - 2022-07-14
+
+### Fixed
+- Segmentation metrics filtering. Prior version artificially boosted performance when filtering was applied.
+
 ## [0.14.7](https://github.com/scaleapi/nucleus-python-client/releases/tag/v0.14.7) - 2022-07-07
 
 ### Added

nucleus/metrics/segmentation_metrics.py

@@ -219,25 +219,21 @@ def _is_instance_segmentation(self, annotation, prediction):
 
     def _filter_confusion_matrix(self, confusion, annotation, prediction):
         if self.annotation_filters or self.prediction_filters:
+            new_confusion = np.zeros_like(confusion)
             # we mask the confusion matrix instead of the images
             if self.annotation_filters:
                 annotation_indexes = {
                     segment.index for segment in annotation.annotations
                 }
-                indexes_to_remove = (
-                    set(range(confusion.shape[0] - 1)) - annotation_indexes
-                )
-                for row in indexes_to_remove:
-                    confusion[row, :] = 0
+                for row in annotation_indexes:
+                    new_confusion[row, :] = confusion[row, :]
             if self.prediction_filters:
                 prediction_indexes = {
                     segment.index for segment in prediction.annotations
                 }
-                indexes_to_remove = (
-                    set(range(confusion.shape[0] - 1)) - prediction_indexes
-                )
-                for col in indexes_to_remove:
-                    confusion[:, col] = 0
+                for col in prediction_indexes:
+                    new_confusion[:, col] = confusion[:, col]
+            confusion = new_confusion
         return confusion
 
 

pyproject.toml

@@ -21,7 +21,7 @@ exclude = '''
 
 [tool.poetry]
 name = "scale-nucleus"
-version = "0.14.7"
+version = "0.14.8"
 description = "The official Python client library for Nucleus, the Data Platform for AI"
 license =  "MIT"
 authors = ["Scale AI Nucleus Team <nucleusapi@scaleapi.com>"]