diff --git a/modules/dnn/src/onnx/onnx_importer.cpp b/modules/dnn/src/onnx/onnx_importer.cpp
index 7421fdbc28..2e2a1ff07f 100644
--- a/modules/dnn/src/onnx/onnx_importer.cpp
+++ b/modules/dnn/src/onnx/onnx_importer.cpp
@@ -2565,8 +2565,11 @@ void ONNXImporter::parseExpand(LayerParams& layerParams, const opencv_onnx::Node
         node_proto.set_input(0, constParams.name);
         node_proto.set_input(1, srcName);
     }
-    else if (broadcast_axes.size() == 1 && broadcast_axes[0] <= 1)
+    else if (broadcast_axes.size() == 1)
     {
+        // FIXME: this will end up creating massive amount of Identity nodes for broadcasting,
+        //        for example, broadcast 1 to 256 needs 256 Identity nodes and 1 Concat node.
+        //        Possible improvement is to use "Scale".
         expandMid(layerParams.name, node_proto, srcName, targetShape[broadcast_axes[0]]);
 
         layerParams.set("axis", broadcast_axes[0]);
@@ -2638,7 +2641,8 @@ void ONNXImporter::parsePad(LayerParams& layerParams, const opencv_onnx::NodePro
         paddings = paddings.t();
         layerParams.set("paddings", DictValue::arrayInt(paddings.ptr<int>(), paddings.total()));
 
-        if (node_proto.input_size() == 3)
+        // check for non-null constant_value
+        if (node_proto.input_size() == 3 && !node_proto.input(2).empty())
         {
             Mat value = getBlob(node_proto, 2);
             float padValue = (depth == CV_8S) ? (float)value.ptr<int8_t>()[0] : value.ptr<float>()[0];