Sorting support for Object Detection labels (#2376)

* sort support ckpt * build fix * python lint fixes * mock data fix
microsoft · Oct 2, 2023 · 896237b · 896237b
1 parent 7206fe1
commit 896237b
Show file tree

Hide file tree

Showing 4 changed files with 17 additions and 10 deletions.
diff --git a/apps/dashboard/src/model-assessment-vision/__mock_data__/fridgeObjectDetection.ts b/apps/dashboard/src/model-assessment-vision/__mock_data__/fridgeObjectDetection.ts
@@ -102,12 +102,12 @@ export const fridgeObjectDetection: IDataset = {
   object_detection_labels: [
     {
       aggregate: "2 correct, 0 incorrect",
-      correct: "1 milk_bottle, 1 can",
+      correct: "1 can, 1 milk_bottle",
       incorrect: "(none)"
     },
     {
       aggregate: "2 correct, 0 incorrect",
-      correct: "1 milk_bottle, 1 can",
+      correct: "1 can, 1 milk_bottle",
       incorrect: "(none)"
     },
     {

diff --git a/libs/core-ui/src/lib/DatasetCohort.ts b/libs/core-ui/src/lib/DatasetCohort.ts
@@ -141,7 +141,7 @@ export class DatasetCohort {
         dataDict[index][DatasetCohortColumns.PredictedY] = val;
       }
     });
-    this.dataset.objectDetectionLabels?.forEach((val, index) => {
+    this.dataset.object_detection_labels?.forEach((val, index) => {
       dataDict[index][DatasetCohortColumns.ObjectDetectionIncorrect] =
         val.incorrect;
       dataDict[index][DatasetCohortColumns.ObjectDetectionCorrect] =

diff --git a/.../model-assessment/src/lib/ModelAssessmentDashboard/Context/buildModelAssessmentContext.ts b/.../model-assessment/src/lib/ModelAssessmentDashboard/Context/buildModelAssessmentContext.ts
@@ -60,7 +60,7 @@ export function buildInitialModelAssessmentContext(
     featureMetaData: props.dataset.feature_metadata,
     localExplanations,
     metadata: modelMetadata,
-    objectDetectionLabels: props.dataset.objectDetectionLabels,
+    objectDetectionLabels: props.dataset.object_detection_labels,
     predictedProbabilities: props.dataset.probability_y,
     predictedY: props.dataset.predicted_y,
     targetColumn: props.dataset.target_column,

diff --git a/responsibleai_vision/responsibleai_vision/rai_vision_insights/rai_vision_insights.py b/responsibleai_vision/responsibleai_vision/rai_vision_insights/rai_vision_insights.py
@@ -738,19 +738,26 @@ def _generate_od_error_labels(self, true_y, pred_y, class_names):
                 else:
                     image_labels[_INCORRECT][object_label] += 1
 
-                image_labels[_INCORRECT][object_label] += \
-                    np.count_nonzero(
-                        error_matrix[label_idx] ==
-                        ErrorLabelType.DUPLICATE_DETECTION)
+                duplicate_detections = np.count_nonzero(
+                    error_matrix[label_idx] ==
+                    ErrorLabelType.DUPLICATE_DETECTION)
+                if duplicate_detections > 0:
+                    image_labels[_INCORRECT][object_label] += \
+                        duplicate_detections
+
+            correct_labels = sorted(image_labels[_CORRECT].items(),
+                                    key=lambda x: class_names.index(x[0]))
+            incorrect_labels = sorted(image_labels[_INCORRECT].items(),
+                                      key=lambda x: class_names.index(x[0]))
 
             rendered_labels[_CORRECT] = ', '.join(
                 f'{value} {key}' for key, value in
-                image_labels[_CORRECT].items() if value > 0)
+                correct_labels)
             if len(rendered_labels[_CORRECT]) == 0:
                 rendered_labels[_CORRECT] = _NOLABEL
             rendered_labels[_INCORRECT] = ', '.join(
                 f'{value} {key}' for key, value in
-                image_labels[_INCORRECT].items() if value > 0)
+                incorrect_labels)
             if len(rendered_labels[_INCORRECT]) == 0:
                 rendered_labels[_INCORRECT] = _NOLABEL
             rendered_labels[_AGGREGATE_LABEL] = \