update preprocess for pipeline (#1240)

kendryte · Nov 8, 2024 · b901f7e · b901f7e
1 parent ed284a2
commit b901f7e
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 5 deletions.
diff --git a/docs/USAGE_v2.md b/docs/USAGE_v2.md
@@ -161,6 +161,9 @@ subgraph A
     end
 
 ```
+
+>如果你在编译`kmodel`时使用了前处理相关的配置，当你需要使用`ONNX`或者`TFLite`框架进行结果验证时，需要在你的`ONNX`或者`TFLite` pipeline中增加相应的前处理操作，这样才能保证和`kmodel` pipeline是等价的。
+
 ##### 动态shape参数
 详见[动态shape参数说明](./shape_bucket.md)
 

diff --git a/docs/USAGE_v2_EN.md b/docs/USAGE_v2_EN.md
@@ -163,8 +163,11 @@ subgraph A
 
 ```
 
+> If you have utilized pre-processing configurations when compiling the `kmodel`, when you need to verify the results using the `ONNX` or `TFLite` framework, you must add the corresponding pre-processing operations to your `ONNX` or `TFLite` pipeline to ensure equivalence between the `kmodel` pipeline.
+
+
 ##### Dynamice shape args
-Refer to[Dynamic shape args description](./shape_bucket.md)
+Refer to [Dynamic shape args description](./shape_bucket.md)
 
 #### Example
 

diff --git a/examples/user_guide/k230_simulate-EN.ipynb b/examples/user_guide/k230_simulate-EN.ipynb
@@ -257,7 +257,9 @@
    "source": [
     "# 5. Compare kmodel result and tflite result.\n",
     "\n",
-    "Here, we will use the TensorFlow framework to infer model(`.tflite`, not kmodel). And calculate the cosine between the tflite result and kmodel result."
+    "Here, we will use the TensorFlow framework to infer model(`.tflite`, not kmodel). And calculate the cosine between the tflite result and kmodel result.\n",
+    "\n",
+    "> If you have utilized pre-processing configurations when compiling the `kmodel`, when you need to verify the results using the `ONNX` or `TFLite` framework, you must add the corresponding pre-processing operations to your `ONNX` or `TFLite` pipeline to ensure equivalence between the `kmodel` pipeline."
    ]
   },
   {
@@ -385,7 +387,9 @@
    "source": [
     "# 8. Compare kmodel results and onnx results.\n",
     "\n",
-    "Here, we will use the ONNX framework to infer model(`.onnx`, not kmodel). And calculate the cosine between the ONNX result and the kmodel result."
+    "Here, we will use the ONNX framework to infer model(`.onnx`, not kmodel). And calculate the cosine between the ONNX result and the kmodel result.\n",
+    "\n",
+    "> If you have utilized pre-processing configurations when compiling the `kmodel`, when you need to verify the results using the `ONNX` or `TFLite` framework, you must add the corresponding pre-processing operations to your `ONNX` or `TFLite` pipeline to ensure equivalence between the `kmodel` pipeline."
    ]
   },
   {

diff --git a/examples/user_guide/k230_simulate-ZH.ipynb b/examples/user_guide/k230_simulate-ZH.ipynb
@@ -257,7 +257,9 @@
    "source": [
     "# 5. 比较TF结果和kmodel的推理结果\n",
     "\n",
-    "这里，我们使用TensorFlow框架来推理`.tflite`模型，然后计算TensorFlow输入结果和kmodel的输出结果的余弦。"
+    "这里，我们使用TensorFlow框架来推理`.tflite`模型，然后计算TensorFlow输入结果和kmodel的输出结果的余弦。\n",
+    "\n",
+    ">如果你在编译`kmodel`时使用了前处理相关的配置，当你需要使用`ONNX`或者`TFLite`框架进行结果验证时，需要在你的`ONNX`或者`TFLite` pipeline中增加相应的前处理操作，这样才能保证和`kmodel` pipeline是等价的。"
    ]
   },
   {
@@ -386,7 +388,9 @@
    "source": [
     "# 8. 比较ONNX结果和kmodel推理结果\n",
     "\n",
-    "这里给出如何调用ONNX框架推理的示例代码，以及如何将ONNX的推理结果和kmodel的推理结果进行比较，该比较结果为余弦值的形式。<br>"
+    "这里给出如何调用ONNX框架推理的示例代码，以及如何将ONNX的推理结果和kmodel的推理结果进行比较，该比较结果为余弦值的形式。\n",
+    "\n",
+    "> 如果你在编译`kmodel`时使用了前处理相关的配置，当你需要使用`ONNX`或者`TFLite`框架进行结果验证时，需要在你的`ONNX`或者`TFLite` pipeline中增加相应的前处理操作，这样才能保证和`kmodel` pipeline是等价的。"
    ]
   },
   {