Fixing detection saliency map for one class case (#2368)

* fix softmax

* fix validity tests

src/otx/algorithms/detection/adapters/mmdet/hooks/det_class_probability_map_hook.py

@@ -62,7 +62,7 @@ def func(
 
         # Don't use softmax for tiles in tiling detection, if the tile doesn't contain objects,
         # it would highlight one of the class maps as a background class
-        if self.use_cls_softmax:
+        if self.use_cls_softmax and self._num_cls_out_channels > 1:
             cls_scores = [torch.softmax(t, dim=1) for t in cls_scores]
 
         batch_size, _, height, width = cls_scores[-1].size()

tests/unit/algorithms/classification/test_xai_classification_validity.py

@@ -54,4 +54,6 @@ def test_saliency_map_cls(self, template):
         assert len(saliency_maps) == 2
         assert saliency_maps[0].ndim == 3
         assert saliency_maps[0].shape == (1000, 7, 7)
-        assert np.all(np.abs(saliency_maps[0][0][0] - self.ref_saliency_vals_cls[template.name]) <= 1)
+        actual_sal_vals = saliency_maps[0][0][0].astype(np.int8)
+        ref_sal_vals = self.ref_saliency_vals_cls[template.name].astype(np.int8)
+        assert np.all(np.abs(actual_sal_vals - ref_sal_vals) <= 1)

tests/unit/algorithms/detection/test_xai_detection_validity.py

@@ -80,7 +80,9 @@ def test_saliency_map_det(self, template):
         assert len(saliency_maps) == 2
         assert saliency_maps[0].ndim == 3
         assert saliency_maps[0].shape == self.ref_saliency_shapes[template.name]
-        assert np.all(np.abs(saliency_maps[0][0][0] - self.ref_saliency_vals_det[template.name]) <= 1)
+        actual_sal_vals = saliency_maps[0][0][0].astype(np.int8)
+        ref_sal_vals = self.ref_saliency_vals_det[template.name].astype(np.int8)
+        assert np.all(np.abs(actual_sal_vals - ref_sal_vals) <= 1)
 
     @e2e_pytest_unit
     @pytest.mark.parametrize("template", templates_det, ids=templates_det_ids)