diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/PyTorch\347\246\273\347\272\277\346\216\250\347\220\206-FAQ.md" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/PyTorch\347\246\273\347\272\277\346\216\250\347\220\206-FAQ.md"
index 614d750dea40ce89f72e936ce7aaf53db9396644..e74e7ae560fee7f3a0cbe35e285cb0a56a717648 100644
--- "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/PyTorch\347\246\273\347\272\277\346\216\250\347\220\206-FAQ.md"
+++ "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/PyTorch\347\246\273\347\272\277\346\216\250\347\220\206-FAQ.md"
@@ -4,9 +4,11 @@
 - [2 om模型转换问题汇总](#2-om模型转换问题汇总)
   - [2.1 如何查看 `ONNX/om/pbtxt` 模型](#21-如何查看-onnxompbtxt-模型)
   - [2.2 `Exporting the operator {opname} to ONNX opset version {version} is not supported.`](#22-exporting-the-operator-opname-to-onnx-opset-version-version-is-not-supported)
+  - [2.3 resize不支持5d输入的解决方案](#23-resize不支持5d输入的解决方案)
 - [3 OM离线推理失败问题汇总](#3-om离线推理失败问题汇总)
   - [3.1 找不到atc命令或找不到ascend动态库](#31-找不到atc命令或找不到ascend动态库)
-  - [模型推理工具常见的错误&&解决方案](#模型推理工具常见的错误解决方案)
+  - [3.2 模型推理工具常见的错误&&解决方案](#32-模型推理工具常见的错误解决方案)
+  - [3.3 msame和benchmark在多batch下推理区别](#33-msame和benchmark在多batch下推理区别)
 - [4 精度调试常见问题](#4-精度调试常见问题)
 - [5 性能优化常见问题](#5-性能优化常见问题)
   - [5.1 如何使用AIPP进行性能提升](#51-如何使用aipp进行性能提升)
@@ -45,6 +47,68 @@
   #x = F.adaptive_avg_pool2d(input, output_size=bin_size)
   x = adaptive_avg_pool_op(input, (bin_size, bin_size))  # 替换上面代码
   ```
+## 2.3 resize不支持5d输入的解决方案
+- 错误现象
+
+  atc转om报错，经定位发现是resize输入是5D的导致后面算子shape不对
+  ```shell
+  # 报错信息
+  E89999: Inner Error!
+          op[Concat_114], the input shape dims should be equal except merge axis.[FUNC:ConcatInferShapeCommon][FILE:split_combination_ops.cc][LINE:705]
+          Call InferShapeAndType for node:Concat_114(ConcatD) failed[FUNC:Infer][FILE:infershape_pass.cc][LINE:117]
+          process pass InferShapePass on node:Concat_114 failed, ret:4294967295[FUNC:RunPassesOnNode][FILE:base_pass.cc][LINE:433]
+          build graph failed, graph id:0, ret:1343242270[FUNC:BuildModel][FILE:ge_generator.cc][LINE:1327]
+  ```
+- 原因分析
+  
+  CANN层目前还不支持5d输入的resize，所以需要改图规避支持。修改前后对比如下
+  ![FAQ001](./images/FAQ001.png)
+- 解决方案
+
+  使用 [MagicONNX工具](https://gitee.com/Ronnie_zheng/MagicONNX/tree/master) 进行改图，这里给出该问题的改图代码，更多功能详见[使用教程](https://gitee.com/Ronnie_zheng/MagicONNX/blob/master/docs/tutorials.md)和[API说明](https://gitee.com/Ronnie_zheng/MagicONNX/blob/master/docs/operations.md)
+  ```python
+  import numpy as np
+  from magiconnx import OnnxGraph
+
+
+  def modify(path):
+      graph = OnnxGraph(path)
+      resizes = graph.get_nodes("Resize")
+      shapes = [[[1, 512*4, 4, 8], [1, 512*4, 8, 16], [1, 512, 4, 8*16], [1, 512, 8, 8*16], [1, 512, 8, 8, 16]],
+                [[1, 256*8, 8, 16], [1, 256*8, 16, 32], [1, 256, 8, 16*32], [1, 256, 16, 16*32], [1, 256, 16, 16, 32]],
+                [[1, 128*16, 16, 32], [1, 128*16, 32, 64], [1, 128, 16, 32*64], [1, 128, 32, 32*64], [1, 128, 32, 32, 64]]]
+      for idx, node in enumerate(resizes):
+          reshape1 = graph.add_node(f'Reshape_{node.name}', 'Reshape')
+          graph.add_initializer(f'shape_{node.name}', np.array(shapes[idx][0]))
+          reshape1.inputs = [node.inputs[0], f'shape_{node.name}']
+          reshape1.outputs = [f'Reshape_{node.name}']
+
+          graph[node.inputs[-1]].value = np.array(shapes[idx][1])
+          out_name = node.outputs[0]
+          node.set_input(0, f'Reshape_{node.name}')
+          node.set_output(0, f'{node.name}_reshape')
+
+          reshape2 = graph.add_node(f'Reshape2_{node.name}', 'Reshape')
+          graph.add_initializer(f'shape2_{node.name}', np.array(shapes[idx][2]))
+          reshape2.inputs = [f'{node.name}_reshape', f'shape2_{node.name}']
+          reshape2.outputs = [f'Reshape2_{node.name}_out']
+
+          resize2 = graph.add_node(f'Resize2_{node.name}', 'Resize')
+          graph.add_initializer(f'size_{node.name}', np.array(shapes[idx][3]))
+          resize2.inputs = [f'Reshape2_{node.name}_out', node.inputs[1], node.inputs[1], f'size_{node.name}']
+          resize2.outputs = [f'Resize2_{node.name}']
+
+          reshape3 = graph.add_node(f'Reshape3_{node.name}', 'Reshape')
+          graph.add_initializer(f'shape3_{node.name}', np.array(shapes[idx][4]))
+          reshape3.inputs = [f'Resize2_{node.name}', f'shape3_{node.name}']
+          reshape3.outputs = [out_name]
+
+      graph.save('modify.onnx')
+
+  if __name__ == "__main__":
+      modify('src.onnx')
+  ```
+
 # 3 OM离线推理失败问题汇总
 ## 3.1 找不到atc命令或找不到ascend动态库
 - 现象描述
@@ -68,7 +132,7 @@
   export ASCEND_AICPU_PATH=/usr/local/Ascend/ascend-toolkit/latest
   ```
 
-## 模型推理工具常见的错误&&解决方案
+## 3.2 模型推理工具常见的错误&&解决方案
 当前默认的模型推理是benchmark工具，以下为常见错误和一些解决方案：
 
 - 错误现象：
@@ -86,6 +150,30 @@
   - 通常是输入的input_image_path/input_text_path格式问题，如常见的nlp模型通常会有多输入的场合会有输入顺序/输入名的问题
   - 当前benchmark工具支持的模型类型有：图像/自然语音/YOLO检测/搜索/语义理解/翻译，但存在不支持的输入类型如：3D的输入（如视频理解/点云等），可以采用[msame工具](https://gitee.com/ascend/tools/tree/master/msame)进行推理。
 
+## 3.3 msame和benchmark在多batch下推理区别
+- 错误现象
+
+  使用msame推理报如下错误
+  ```shell
+  [ERROR] GE(7672,msame):2021-11-16-12:57:02.335.376 [davinci_model.cc:3466]7672 CheckUserAndModelSize: ErrorNo: 145000(Parameter invalid.) [EXEC][DEFAULT][Check][Param] input size:150528 from user add align:64 < op_size:2408448 in model, model_id:1
+  [ERROR] GE(7672,msame):2021-11-16-12:57:02.335.386 [davinci_model.cc:3546]7672 UpdateIoTaskArgs: ErrorNo: 145000(Parameter invalid.) [EXEC][DEFAULT][Call][CheckInputAndModelSize] failed, op[image]
+  [ERROR] GE(7672,msame):2021-11-16-12:57:02.335.395 [davinci_model.cc:3483]7672 CopyModelData: ErrorNo: 145000(Parameter invalid.) [EXEC][DEFAULT][Call][UpdateIoTaskArgs] [ZCPY] Update input data to model:1 failed.
+  [ERROR] GE(7672,msame):2021-11-16-12:57:02.335.406 [davinci_model.cc:3921]7672 NnExecute: ErrorNo: 4294967295(failed) [EXEC][DEFAULT][Copy][ModelData] failed. model id: 1
+  [ERROR] GE(7672,msame):2021-11-16-12:57:02.335.414 [graph_loader.cc:283]7672 ExecuteModel: ErrorNo: 145000(Parameter invalid.) [EXEC][DEFAULT][Execute][Model] failed, model_id:1.
+  [ERROR] ASCENDCL(7672,msame):2021-11-16-12:57:02.335.428 [model.cpp:824]7672 ModelExecute: [EXEC][DEFAULT][Exec][Model]Execute model failed, ge result[145000], modelId[1]
+  [ERROR] ASCENDCL(7672,msame):2021-11-16-12:57:02.335.446 [model.cpp:869]7672 aclmdlExecute: [EXEC][DEFAULT][Exec][Model]modelId[1] execute failed, result[145000]
+  [ERROR] execute model failed, modelId is 1
+  [ERROR] model execute failed
+  [ERROR] Sample process failed
+  ```
+  - 原因分析
+
+    benchmark工具多batch模式下设置 `batch_size=16` 数值后，benchmark工具会一次读取16个bin进行推理，所以一个bin对应的是一张图片；
+    msame工具只会一次读取1个bin进行推理（没有`batch_size`命令行参数），所以多batch模式下需要一个bin对应的是16张图片；
+  - 解决方案
+
+    预处理时把16张图片保存成一个bin文件
+
 # 4 精度调试常见问题
 
 # 5 性能优化常见问题
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/images/FAQ001.png" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/images/FAQ001.png"
new file mode 100644
index 0000000000000000000000000000000000000000..0c396ece29c3a7daaffe1c0aeba1adfd4a475d5a
Binary files /dev/null and "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/images/FAQ001.png" differ