diff --git a/tools/mlir/mlir2ncnn.cpp b/tools/mlir/mlir2ncnn.cpp
index 51f744edd..20cf277d2 100644
--- a/tools/mlir/mlir2ncnn.cpp
+++ b/tools/mlir/mlir2ncnn.cpp
@@ -420,6 +420,10 @@ int main(int argc, char** argv)
         {
             fprintf(pp, "%-16s", "InstanceNorm");
         }
+        else if (op == "ncnn.InstanceNormAffine")
+        {
+            fprintf(pp, "%-16s", "InstanceNorm");
+        }
         else if (op == "tf.AddN")
         {
             fprintf(pp, "%-16s", "Eltwise");
@@ -617,6 +621,14 @@ int main(int argc, char** argv)
         {
             float eps = get_operation_attr_f(operation, "epsilon");
 
+            fprintf(pp, " 0=0"); // channels
+            fprintf(pp, " 1=%e", eps);
+            fprintf(pp, " 2=0"); // affine
+        }
+        else if (op == "ncnn.InstanceNormAffine")
+        {
+            float eps = get_operation_attr_f(operation, "epsilon");
+
             std::string gamma_name = get_mlir_value_uniq_id(operation.getOperand(1));
             std::string beta_name = get_mlir_value_uniq_id(operation.getOperand(2));
             const mlir::Attribute& G = weights[gamma_name];
@@ -629,6 +641,7 @@ int main(int argc, char** argv)
 
             fprintf(pp, " 0=%d", channels);
             fprintf(pp, " 1=%e", eps);
+            fprintf(pp, " 2=1"); // affine
 
             fwrite(gv.data(), sizeof(float), gv.size(), bp);
             fwrite(bv.data(), sizeof(float), bv.size(), bp);
diff --git a/tools/mlir/ncnn_ops.td b/tools/mlir/ncnn_ops.td
index d71f2856c..086554488 100644
--- a/tools/mlir/ncnn_ops.td
+++ b/tools/mlir/ncnn_ops.td
@@ -36,6 +36,20 @@ class NCNN_Op<string mnemonic, list<OpTrait> traits = []> :
 
 def NCNN_InstanceNormOp : NCNN_Op<"InstanceNorm", [NoSideEffect]> {
 
+  let arguments = (ins
+    F32Tensor:$x,
+    F32Attr:$epsilon
+  );
+
+  let results = (outs
+    F32Tensor:$y
+  );
+
+  let hasCanonicalizer = 1;
+}
+
+def NCNN_InstanceNormAffineOp : NCNN_Op<"InstanceNormAffine", [NoSideEffect]> {
+
   let arguments = (ins
     F32Tensor:$x,
     F32Tensor:$gamma,
diff --git a/tools/mlir/ncnn_rewriter.cpp b/tools/mlir/ncnn_rewriter.cpp
index b733e29ce..97063b701 100644
--- a/tools/mlir/ncnn_rewriter.cpp
+++ b/tools/mlir/ncnn_rewriter.cpp
@@ -30,7 +30,13 @@ namespace ncnn {
 
 void InstanceNormOp::getCanonicalizationPatterns(OwningRewritePatternList& results, MLIRContext* context)
 {
-    results.insert<FuseInstanceNormPattern>(context);
+    results.insert<FuseInstanceNormPattern0>(context);
+    results.insert<FuseInstanceNormPattern1>(context);
+}
+
+void InstanceNormAffineOp::getCanonicalizationPatterns(OwningRewritePatternList& results, MLIRContext* context)
+{
+    results.insert<FuseInstanceNormAffinePattern>(context);
 }
 
 } // namespace ncnn
diff --git a/tools/mlir/ncnn_rewriter.td b/tools/mlir/ncnn_rewriter.td
index c2f123d0e..a9bcdfc41 100644
--- a/tools/mlir/ncnn_rewriter.td
+++ b/tools/mlir/ncnn_rewriter.td
@@ -22,7 +22,71 @@ def get_attr_f : NativeCodeCall<"$0.getValue<FloatAttr>(0)">;
 
 def EqualOperands : Constraint<CPred<"$0 == $1">>;
 
-def FuseInstanceNormPattern : Pat<
+def FuseInstanceNormPattern0 : Pat<
+    (TF_MulOp
+        (TF_RsqrtOp
+            (TF_AddV2Op
+                (TF_MeanOp
+                    (TF_SquaredDifferenceOp
+                        (TF_MeanOp:$mean
+                            $x,
+                            (TF_ConstOp:$reduce_axis ElementsAttr),
+                            ConstBoolAttrTrue // keep_dims
+                        ),
+                        $x_
+                    ),
+                    $reduce_axis_,
+                    ConstBoolAttrTrue // keep_dims
+                ),
+                (TF_ConstOp ElementsAttr:$epsilon)
+            )
+        ),
+        (TF_SubOp $x__, $mean_)
+    ),
+
+    (NCNN_InstanceNormOp $x, (get_attr_f $epsilon)),
+
+    [
+        (EqualOperands $x, $x_),
+        (EqualOperands $x, $x__),
+        (EqualOperands $reduce_axis, $reduce_axis_),
+        (EqualOperands $mean, $mean_)
+    ]
+>;
+
+def FuseInstanceNormPattern1 : Pat<
+    (TF_MulOp
+        (TF_RsqrtOp
+            (TF_AddV2Op
+                (TF_MeanOp
+                    (TF_SquaredDifferenceOp
+                        $x_,
+                        (TF_MeanOp:$mean
+                            $x,
+                            (TF_ConstOp:$reduce_axis ElementsAttr),
+                            ConstBoolAttrTrue // keep_dims
+                        )
+                    ),
+                    $reduce_axis_,
+                    ConstBoolAttrTrue // keep_dims
+                ),
+                (TF_ConstOp ElementsAttr:$epsilon)
+            )
+        ),
+        (TF_SubOp $x__, $mean_)
+    ),
+
+    (NCNN_InstanceNormOp $x, (get_attr_f $epsilon)),
+
+    [
+        (EqualOperands $x, $x_),
+        (EqualOperands $x, $x__),
+        (EqualOperands $reduce_axis, $reduce_axis_),
+        (EqualOperands $mean, $mean_)
+    ]
+>;
+
+def FuseInstanceNormAffinePattern : Pat<
     (TF_ReshapeOp
         (TF_AddV2Op
             (TF_MulOp
@@ -56,7 +120,7 @@ def FuseInstanceNormPattern : Pat<
         (TF_ConstOp ElementsAttr)
     ),
 
-    (NCNN_InstanceNormOp $x, $gamma, $beta, (get_attr_f $epsilon)),
+    (NCNN_InstanceNormAffineOp $x, $gamma, $beta, (get_attr_f $epsilon)),
 
     [
         (EqualOperands $reshaped, $reshaped_),