[Benchmark]Compare diff for OCR (#1415)

* avoid mem copy for cpp benchmark

* set CMAKE_BUILD_TYPE to Release

* Add SegmentationDiff

* change pointer to reference

* fixed bug

* cast uint8 to int32

* Add diff compare for OCR

* Add diff compare for OCR

* rm ppocr pipeline

* Add yolov5 diff compare

* Add yolov5 diff compare

* deal with comments

* deal with comments

* fixed bug

* fixed bug

fastdeploy/vision/facealign/contrib/pipnet.cc

@@ -632,7 +632,7 @@ bool PIPNet::Preprocess(Mat* mat, FDTensor* output,
   HWC2CHW::Run(mat);
   Cast::Run(mat, "float");
   mat->ShareWithTensor(output);
-  output->shape.insert(output->shape.begin(), 1);  // reshape to n, h, w, c
+  output->shape.insert(output->shape.begin(), 1);  // reshape to n, c, h, w
   return true;
 }