Fixing graph visualization

src/data_gradients/config/detection.yaml

@@ -22,3 +22,4 @@ report_sections:
       - DetectionClassesPerImageCount
       - DetectionBoundingBoxIoU:
           num_bins: 10
+          class_agnostic: true

src/data_gradients/feature_extractors/object_detection/bounding_boxes_area.py

@@ -31,6 +31,7 @@ def update(self, sample: DetectionSample):
     def aggregate(self) -> Feature:
         df = pd.DataFrame(self.data)
 
+        max_area = min(100, df["relative_bbox_area"].max())
         plot_options = ViolinPlotOptions(
             x_label_key="relative_bbox_area",
             x_label_name="Bounding Box Area (in % of image)",
@@ -40,6 +41,7 @@ def aggregate(self) -> Feature:
             title=self.title,
             x_ticks_rotation=None,
             labels_key="split",
+            x_lim=(0, max_area),
             bandwidth=0.4,
         )
 

src/data_gradients/feature_extractors/object_detection/bounding_boxes_iou.py

@@ -17,7 +17,7 @@ class DetectionBoundingBoxIoU(AbstractFeatureExtractor):
     """Feature Extractor to compute the pairwise IoU of bounding boxes per image.
     This feature extractor helps to identify duplicate/highly overlapping bounding boxes."""
 
-    def __init__(self, num_bins: int = 10, class_agnostic: bool = False):
+    def __init__(self, num_bins: int = 10, class_agnostic: bool = True):
         """
         :param num_bins: Number of bins to use for the heatmap plot.
         :param class_agnostic: If True, only check IoU of bounding boxes of the same class.

src/data_gradients/feature_extractors/object_detection/bounding_boxes_per_image_count.py

@@ -37,6 +37,7 @@ def aggregate(self) -> Feature:
             kde=False,
             labels_key="split",
             individual_plots_key="split",
+            stat="percent",
             x_ticks_rotation=None,
             sharey=True,
             labels_palette=LABELS_PALETTE,

src/data_gradients/feature_extractors/object_detection/classes_frequency_per_image.py

@@ -42,6 +42,7 @@ def aggregate(self) -> Feature:
             y_label_name="Class Names",
             order_key="class_id",
             title=self.title,
+            x_lim=(0, df_class_count["n_appearance"].max() * 1.2),
             bandwidth=0.4,
             x_ticks_rotation=None,
             labels_key="split",