fix: dense layer filtering now works properly

rfa_toolbox/utils/graph_utils.py

@@ -106,7 +106,16 @@ def filters_non_convolutional_node(
         are treated like a convolutional layer with a kernel
         and stride size of 1.
     """
-    return [node for node in nodes if node.layer_info.kernel_size != np.inf]
+    result = []
+    for node in nodes:
+        if isinstance(node.receptive_field_min, Sequence) or isinstance(
+            node.receptive_field_min, tuple
+        ):
+            if not np.any(np.isinf(node.receptive_field_min)):
+                result.append(node)
+        elif node.receptive_field_min != np.inf:
+            result.append(node)
+    return result
 
 
 def input_resolution_range(