Added support for constant folding Quant weight nodes with per-channel scaling for convolutions.

src/finn/transformation/convert_qonnx_to_finn.py

@@ -109,6 +109,21 @@ class FoldQuantWeights(Transformation):
                         new_dtype = DataType[dtype.name.replace("SCALED", "")]
                         model.set_tensor_datatype(node_out, new_dtype)
 
+                        # Reshape scale for Conv if required
+                        target_node = model.find_direct_successors(n)
+                        if target_node is None:
+                            raise RuntimeError(
+                                "Weights quantized with the Quant node must have "
+                                "a successor node."
+                            )
+                        else:
+                            target_node = target_node[0]
+
+                        if target_node.op_type == "Conv" and len(scale.shape) > 0:
+                            bias_shape = [1] * len(scale.shape)
+                            bias_shape[1] = -1
+                            scale = scale.reshape(bias_shape)
+
                         if scale.shape == (1,):
                             scale = scale[0]
                             mul_shape = tuple()