fix bbox bug and tests

openvino_xai/methods/black_box/aise.py

@@ -372,12 +372,15 @@ def generate_saliency_map(  # type: ignore
 
         saliency_maps = {}
         for target in target_indices:
-            self.kernel_params_hist = collections.defaultdict(list)
-            self.pred_score_hist = collections.defaultdict(list)
-
             self.target_box = boxes[target]
             self.target_label = labels[target]
 
+            if self.target_box[0] == self.target_box[2] or self.target_box[1] == self.target_box[3]:
+                continue
+
+            self.kernel_params_hist = collections.defaultdict(list)
+            self.pred_score_hist = collections.defaultdict(list)
+
             self._process_box()
             saliency_map_per_target = self._run_synchronous_explanation()
             if scale_output:
@@ -427,6 +430,10 @@ def _process_box(self, padding_coef: float = 0.5) -> None:
         x_to = min(target_box_scaled[2] + box_width * padding_coef, 1.0)
         y_from = max(target_box_scaled[1] - box_height * padding_coef, 0.0)
         y_to = min(target_box_scaled[3] + box_height * padding_coef, 1.0)
+
+        if x_from < x_to or y_from < y_to:
+            raise ValueError("Bounding box data is incorrect.")
+
         self.bounds = Bounds([x_from, y_from], [x_to, y_to])
 
     def _get_loss(self, data_perturbed: np.array) -> float:

tests/unit/explanation/test_visualization.py

@@ -4,13 +4,18 @@
 import numpy as np
 import pytest
 
+from openvino_xai.common.parameters import Task
 from openvino_xai.common.utils import get_min_max, scaling
 from openvino_xai.explainer.explanation import Explanation
 from openvino_xai.explainer.visualizer import Visualizer, colormap, overlay, resize
 
 SALIENCY_MAPS = [
     (np.random.rand(1, 5, 5) * 255).astype(np.uint8),
     (np.random.rand(1, 2, 5, 5) * 255).astype(np.uint8),
+    {
+        0: (np.random.rand(5, 5) * 255).astype(np.uint8),
+        1: (np.random.rand(5, 5) * 255).astype(np.uint8),
+    },
 ]
 
 EXPLAIN_ALL_CLASSES = [
@@ -98,7 +103,7 @@ class TestVisualizer:
     @pytest.mark.parametrize("colormap", [True, False])
     @pytest.mark.parametrize("overlay", [True, False])
     @pytest.mark.parametrize("overlay_weight", [0.5, 0.3])
-    def test_Visualizer(
+    def test_visualizer(
         self,
         saliency_maps,
         explain_all_classes,
@@ -142,7 +147,7 @@ def test_Visualizer(
             for map_ in explanation.saliency_map.values():
                 assert map_.shape[:2] == original_input_image.shape[:2]
 
-        if saliency_maps.ndim == 3 and not overlay:
+        if isinstance(saliency_maps, np.ndarray) and saliency_maps.ndim == 3 and not overlay:
             explanation = Explanation(saliency_maps, targets=-1)
             visualizer = Visualizer()
             explanation_output_size = visualizer(
@@ -157,3 +162,23 @@ def test_Visualizer(
             maps_data = explanation.saliency_map
             maps_size = explanation_output_size.saliency_map
             assert np.all(maps_data["per_image_map"] == maps_size["per_image_map"])
+
+        if isinstance(saliency_maps, dict):
+            metadata = {
+                Task.DETECTION: {
+                    0: ([5, 0, 7, 4], 0.5, 0),
+                    1: ([2, 5, 9, 7], 0.5, 0),
+                }
+            }
+            explanation = Explanation(saliency_maps, targets=-1, metadata=metadata)
+            visualizer = Visualizer()
+            explanation_output_size = visualizer(
+                explanation=explanation,
+                original_input_image=original_input_image,
+                output_size=(20, 20),
+                scaling=scaling,
+                resize=resize,
+                colormap=colormap,
+                overlay=overlay,
+                overlay_weight=overlay_weight,
+            )

tests/unit/methods/black_box/test_black_box_method.py

@@ -137,12 +137,23 @@ def test_run(self, target_indices, fxt_data_root: Path):
         assert saliency_map[ref_target].shape == (416, 416)
         assert (saliency_map[ref_target] >= 0).all() and (saliency_map[ref_target] <= 255).all()
 
-        tmp = saliency_map[0]
-
         actual_sal_vals = saliency_map[0][150, 240:250].astype(np.int16)
         ref_sal_vals = np.array([152, 168, 184, 199, 213, 225, 235, 243, 247, 249], dtype=np.uint8)
         assert np.all(np.abs(actual_sal_vals - ref_sal_vals) <= 1)
 
+    def test_target_none(self, fxt_data_root: Path):
+        model = self.get_det_model(fxt_data_root)
+
+        aise_method = AISEDetection(model, self.postprocess_det_fn, self.preprocess_det_fn)
+        saliency_map = aise_method.generate_saliency_map(
+            data=self.image,
+            target_indices=None,
+            preset=Preset.SPEED,
+            num_iterations_per_kernel=1,
+            divisors=[5],
+        )
+        assert len(saliency_map) == 56
+
     def test_preset(self, fxt_data_root: Path):
         model = self.get_det_model(fxt_data_root)
         method = AISEDetection(model, self.postprocess_det_fn, self.preprocess_det_fn)