[release 1.5.0] Fix XAI algorithm for Detection (#2617)

Update detection XAI algorithm

Author: Galina Zalesskaya

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

@@ -60,12 +60,9 @@ def func(
         else:
             cls_scores = self._get_cls_scores_from_feature_map(feature_map)
 
-        # 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 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()
+        middle_idx = len(cls_scores) // 2
+        # resize to the middle feature map
+        batch_size, _, height, width = cls_scores[middle_idx].size()
         saliency_maps = torch.empty(batch_size, self._num_cls_out_channels, height, width)
         for batch_idx in range(batch_size):
             cls_scores_anchorless = []
@@ -82,6 +79,11 @@ def func(
                 )
             saliency_maps[batch_idx] = torch.cat(cls_scores_anchorless_resized, dim=0).mean(dim=0)
 
+        # 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:
+            saliency_maps[0] = torch.stack([torch.softmax(t, dim=1) for t in saliency_maps[0]])
+
         if self._norm_saliency_maps:
             saliency_maps = saliency_maps.reshape((batch_size, self._num_cls_out_channels, -1))
             saliency_maps = self._normalize_map(saliency_maps)

tests/unit/algorithms/detection/test_xai_detection_validity.py

@@ -24,31 +24,31 @@
 
 class TestExplainMethods:
     ref_saliency_shapes = {
-        "MobileNetV2-ATSS": (2, 4, 4),
-        "ResNeXt101-ATSS": (2, 4, 4),
+        "MobileNetV2-ATSS": (2, 13, 13),
+        "ResNeXt101-ATSS": (2, 13, 13),
         "SSD": (81, 13, 13),
-        "YOLOX-TINY": (80, 13, 13),
-        "YOLOX-S": (80, 13, 13),
-        "YOLOX-L": (80, 13, 13),
-        "YOLOX-X": (80, 13, 13),
+        "YOLOX-TINY": (80, 26, 26),
+        "YOLOX-S": (80, 26, 26),
+        "YOLOX-L": (80, 26, 26),
+        "YOLOX-X": (80, 26, 26),
     }
 
     ref_saliency_vals_det = {
-        "MobileNetV2-ATSS": np.array([67, 216, 255, 57], dtype=np.uint8),
-        "ResNeXt101-ATSS": np.array([75, 214, 229, 173], dtype=np.uint8),
-        "YOLOX-TINY": np.array([80, 28, 42, 53, 49, 68, 72, 75, 69, 57, 65, 6, 157], dtype=np.uint8),
-        "YOLOX-S": np.array([75, 178, 151, 159, 150, 148, 144, 144, 147, 144, 147, 142, 189], dtype=np.uint8),
-        "YOLOX-L": np.array([43, 28, 0, 6, 7, 19, 22, 17, 14, 18, 25, 7, 34], dtype=np.uint8),
-        "YOLOX-X": np.array([255, 144, 83, 76, 83, 86, 82, 90, 91, 93, 110, 104, 83], dtype=np.uint8),
-        "SSD": np.array([119, 72, 118, 35, 39, 30, 31, 31, 36, 27, 44, 23, 61], dtype=np.uint8),
+        "MobileNetV2-ATSS": np.array([34, 67, 148, 132, 172, 147, 146, 155, 167, 159], dtype=np.uint8),
+        "ResNeXt101-ATSS": np.array([52, 75, 68, 76, 89, 94, 101, 111, 125, 123], dtype=np.uint8),
+        "YOLOX-TINY": np.array([177, 94, 147, 147, 161, 162, 164, 164, 163, 166], dtype=np.uint8),
+        "YOLOX-S": np.array([158, 170, 180, 158, 152, 148, 153, 153, 148, 145], dtype=np.uint8),
+        "YOLOX-L": np.array([255, 80, 97, 88, 73, 71, 72, 76, 75, 76], dtype=np.uint8),
+        "YOLOX-X": np.array([185, 218, 189, 103, 83, 70, 62, 66, 66, 67], dtype=np.uint8),
+        "SSD": np.array([255, 178, 212, 90, 93, 79, 79, 80, 87, 83], dtype=np.uint8),
     }
 
     ref_saliency_vals_det_wo_postprocess = {
-        "MobileNetV2-ATSS": -0.10465062,
-        "ResNeXt101-ATSS": -0.073549636,
+        "MobileNetV2-ATSS": -0.014513552,
+        "ResNeXt101-ATSS": -0.055565584,
         "YOLOX-TINY": 0.04948914,
-        "YOLOX-S": 0.01133332,
-        "YOLOX-L": 0.01870133,
+        "YOLOX-S": 0.011557617,
+        "YOLOX-L": 0.020231,
         "YOLOX-X": 0.0043506604,
         "SSD": 0.6629989,
     }
@@ -93,7 +93,7 @@ def test_saliency_map_det(self, template):
         assert saliency_maps[0].ndim == 3
         assert saliency_maps[0].shape == self.ref_saliency_shapes[template.name]
         # convert to int16 in case of negative value difference
-        actual_sal_vals = saliency_maps[0][0][0].astype(np.int16)
+        actual_sal_vals = saliency_maps[0][0][0][:10].astype(np.int16)
         ref_sal_vals = self.ref_saliency_vals_det[template.name].astype(np.uint8)
         assert np.all(np.abs(actual_sal_vals - ref_sal_vals) <= 1)