提交 297edfaa 编写于 作者: littletomatodonkey's avatar littletomatodonkey

add cpp inference

上级 a987ec70
# Byte-compiled / optimized / DLL files # Byte-compiled / optimized / DLL files
__pycache__/ __pycache__/
.ipynb_checkpoints/
*.py[cod] *.py[cod]
*$py.class *$py.class
......
project(ocr_system CXX C)
option(WITH_MKL "Compile demo with MKL/OpenBlas support, default use MKL." ON)
option(WITH_GPU "Compile demo with GPU/CPU, default use CPU." OFF)
option(WITH_STATIC_LIB "Compile demo with static/shared library, default use static." ON)
option(USE_TENSORRT "Compile demo with TensorRT." OFF)
macro(safe_set_static_flag)
foreach(flag_var
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
if(${flag_var} MATCHES "/MD")
string(REGEX REPLACE "/MD" "/MT" ${flag_var} "${${flag_var}}")
endif(${flag_var} MATCHES "/MD")
endforeach(flag_var)
endmacro()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -g -fpermissive")
set(CMAKE_STATIC_LIBRARY_PREFIX "")
message("flags" ${CMAKE_CXX_FLAGS})
set(CMAKE_CXX_FLAGS_RELEASE "-O3")
if(NOT DEFINED PADDLE_LIB)
message(FATAL_ERROR "please set PADDLE_LIB with -DPADDLE_LIB=/path/paddle/lib")
endif()
if(NOT DEFINED DEMO_NAME)
message(FATAL_ERROR "please set DEMO_NAME with -DDEMO_NAME=demo_name")
endif()
# user ze
# find_package(OpenCV)
set(OPENCV_DIR "/paddle/libs/opencv-3.4.7/opencv3")
find_package(OpenCV REQUIRED PATHS ${OPENCV_DIR}/share/OpenCV NO_DEFAULT_PATH)
include_directories(${OpenCV_INCLUDE_DIRS})
include_directories("${PADDLE_LIB}/paddle/include")
include_directories("${PADDLE_LIB}/third_party/install/protobuf/include")
include_directories("${PADDLE_LIB}/third_party/install/glog/include")
include_directories("${PADDLE_LIB}/third_party/install/gflags/include")
include_directories("${PADDLE_LIB}/third_party/install/xxhash/include")
include_directories("${PADDLE_LIB}/third_party/install/zlib/include")
include_directories("${PADDLE_LIB}/third_party/boost")
include_directories("${PADDLE_LIB}/third_party/eigen3")
include_directories("${CMAKE_SOURCE_DIR}/")
if (USE_TENSORRT AND WITH_GPU)
include_directories("${TENSORRT_ROOT}/include")
link_directories("${TENSORRT_ROOT}/lib")
endif()
link_directories("${PADDLE_LIB}/third_party/install/zlib/lib")
link_directories("${PADDLE_LIB}/third_party/install/protobuf/lib")
link_directories("${PADDLE_LIB}/third_party/install/glog/lib")
link_directories("${PADDLE_LIB}/third_party/install/gflags/lib")
link_directories("${PADDLE_LIB}/third_party/install/xxhash/lib")
link_directories("${PADDLE_LIB}/paddle/lib")
add_executable(${DEMO_NAME} src/main.cpp src/ocr_det.cpp src/ocr_rec.cpp src/preprocess_op.cpp src/clipper.cpp src/postprocess_op.cpp )
if(WITH_MKL)
include_directories("${PADDLE_LIB}/third_party/install/mklml/include")
set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel${CMAKE_SHARED_LIBRARY_SUFFIX}
${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5${CMAKE_SHARED_LIBRARY_SUFFIX})
set(MKLDNN_PATH "${PADDLE_LIB}/third_party/install/mkldnn")
if(EXISTS ${MKLDNN_PATH})
include_directories("${MKLDNN_PATH}/include")
set(MKLDNN_LIB ${MKLDNN_PATH}/lib/libmkldnn.so.0)
endif()
else()
set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas${CMAKE_STATIC_LIBRARY_SUFFIX})
endif()
# Note: libpaddle_inference_api.so/a must put before libpaddle_fluid.so/a
if(WITH_STATIC_LIB)
set(DEPS
${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_STATIC_LIBRARY_SUFFIX})
else()
set(DEPS
${PADDLE_LIB}/paddle/lib/libpaddle_fluid${CMAKE_SHARED_LIBRARY_SUFFIX})
endif()
# user ze
# set(EXTERNAL_LIB "-lrt -ldl -lpthread -lm -lopencv_world")
# gry
set(EXTERNAL_LIB "-lrt -ldl -lpthread -lm")
set(DEPS ${DEPS}
${MATH_LIB} ${MKLDNN_LIB}
glog gflags protobuf z xxhash
${EXTERNAL_LIB} ${OpenCV_LIBS})
if(WITH_GPU)
if (USE_TENSORRT)
set(DEPS ${DEPS}
${TENSORRT_ROOT}/lib/libnvinfer${CMAKE_SHARED_LIBRARY_SUFFIX})
set(DEPS ${DEPS}
${TENSORRT_ROOT}/lib/libnvinfer_plugin${CMAKE_SHARED_LIBRARY_SUFFIX})
endif()
set(DEPS ${DEPS} ${CUDA_LIB}/libcudart${CMAKE_SHARED_LIBRARY_SUFFIX})
set(DEPS ${DEPS} ${CUDA_LIB}/libcudart${CMAKE_SHARED_LIBRARY_SUFFIX} )
set(DEPS ${DEPS} ${CUDA_LIB}/libcublas${CMAKE_SHARED_LIBRARY_SUFFIX} )
set(DEPS ${DEPS} ${CUDNN_LIB}/libcudnn${CMAKE_SHARED_LIBRARY_SUFFIX} )
endif()
target_link_libraries(${DEMO_NAME} ${DEPS})
/*******************************************************************************
* *
* Author : Angus Johnson *
* Version : 6.4.2 *
* Date : 27 February 2017 *
* Website : http://www.angusj.com *
* Copyright : Angus Johnson 2010-2017 *
* *
* License: *
* Use, modification & distribution is subject to Boost Software License Ver 1. *
* http://www.boost.org/LICENSE_1_0.txt *
* *
* Attributions: *
* The code in this library is an extension of Bala Vatti's clipping algorithm: *
* "A generic solution to polygon clipping" *
* Communications of the ACM, Vol 35, Issue 7 (July 1992) pp 56-63. *
* http://portal.acm.org/citation.cfm?id=129906 *
* *
* Computer graphics and geometric modeling: implementation and algorithms *
* By Max K. Agoston *
* Springer; 1 edition (January 4, 2005) *
* http://books.google.com/books?q=vatti+clipping+agoston *
* *
* See also: *
* "Polygon Offsetting by Computing Winding Numbers" *
* Paper no. DETC2005-85513 pp. 565-575 *
* ASME 2005 International Design Engineering Technical Conferences *
* and Computers and Information in Engineering Conference (IDETC/CIE2005) *
* September 24-28, 2005 , Long Beach, California, USA *
* http://www.me.berkeley.edu/~mcmains/pubs/DAC05OffsetPolygon.pdf *
* *
*******************************************************************************/
#ifndef clipper_hpp
#define clipper_hpp
#define CLIPPER_VERSION "6.4.2"
// use_int32: When enabled 32bit ints are used instead of 64bit ints. This
// improve performance but coordinate values are limited to the range +/- 46340
//#define use_int32
// use_xyz: adds a Z member to IntPoint. Adds a minor cost to perfomance.
//#define use_xyz
// use_lines: Enables line clipping. Adds a very minor cost to performance.
#define use_lines
// use_deprecated: Enables temporary support for the obsolete functions
//#define use_deprecated
#include <cstdlib>
#include <cstring>
#include <functional>
#include <list>
#include <ostream>
#include <queue>
#include <set>
#include <stdexcept>
#include <vector>
namespace ClipperLib {
enum ClipType { ctIntersection, ctUnion, ctDifference, ctXor };
enum PolyType { ptSubject, ptClip };
// By far the most widely used winding rules for polygon filling are
// EvenOdd & NonZero (GDI, GDI+, XLib, OpenGL, Cairo, AGG, Quartz, SVG, Gr32)
// Others rules include Positive, Negative and ABS_GTR_EQ_TWO (only in OpenGL)
// see http://glprogramming.com/red/chapter11.html
enum PolyFillType { pftEvenOdd, pftNonZero, pftPositive, pftNegative };
#ifdef use_int32
typedef int cInt;
static cInt const loRange = 0x7FFF;
static cInt const hiRange = 0x7FFF;
#else
typedef signed long long cInt;
static cInt const loRange = 0x3FFFFFFF;
static cInt const hiRange = 0x3FFFFFFFFFFFFFFFLL;
typedef signed long long long64; // used by Int128 class
typedef unsigned long long ulong64;
#endif
struct IntPoint {
cInt X;
cInt Y;
#ifdef use_xyz
cInt Z;
IntPoint(cInt x = 0, cInt y = 0, cInt z = 0) : X(x), Y(y), Z(z){};
#else
IntPoint(cInt x = 0, cInt y = 0) : X(x), Y(y){};
#endif
friend inline bool operator==(const IntPoint &a, const IntPoint &b) {
return a.X == b.X && a.Y == b.Y;
}
friend inline bool operator!=(const IntPoint &a, const IntPoint &b) {
return a.X != b.X || a.Y != b.Y;
}
};
//------------------------------------------------------------------------------
typedef std::vector<IntPoint> Path;
typedef std::vector<Path> Paths;
inline Path &operator<<(Path &poly, const IntPoint &p) {
poly.push_back(p);
return poly;
}
inline Paths &operator<<(Paths &polys, const Path &p) {
polys.push_back(p);
return polys;
}
std::ostream &operator<<(std::ostream &s, const IntPoint &p);
std::ostream &operator<<(std::ostream &s, const Path &p);
std::ostream &operator<<(std::ostream &s, const Paths &p);
struct DoublePoint {
double X;
double Y;
DoublePoint(double x = 0, double y = 0) : X(x), Y(y) {}
DoublePoint(IntPoint ip) : X((double)ip.X), Y((double)ip.Y) {}
};
//------------------------------------------------------------------------------
#ifdef use_xyz
typedef void (*ZFillCallback)(IntPoint &e1bot, IntPoint &e1top, IntPoint &e2bot,
IntPoint &e2top, IntPoint &pt);
#endif
enum InitOptions {
ioReverseSolution = 1,
ioStrictlySimple = 2,
ioPreserveCollinear = 4
};
enum JoinType { jtSquare, jtRound, jtMiter };
enum EndType {
etClosedPolygon,
etClosedLine,
etOpenButt,
etOpenSquare,
etOpenRound
};
class PolyNode;
typedef std::vector<PolyNode *> PolyNodes;
class PolyNode {
public:
PolyNode();
virtual ~PolyNode(){};
Path Contour;
PolyNodes Childs;
PolyNode *Parent;
PolyNode *GetNext() const;
bool IsHole() const;
bool IsOpen() const;
int ChildCount() const;
private:
// PolyNode& operator =(PolyNode& other);
unsigned Index; // node index in Parent.Childs
bool m_IsOpen;
JoinType m_jointype;
EndType m_endtype;
PolyNode *GetNextSiblingUp() const;
void AddChild(PolyNode &child);
friend class Clipper; // to access Index
friend class ClipperOffset;
};
class PolyTree : public PolyNode {
public:
~PolyTree() { Clear(); };
PolyNode *GetFirst() const;
void Clear();
int Total() const;
private:
// PolyTree& operator =(PolyTree& other);
PolyNodes AllNodes;
friend class Clipper; // to access AllNodes
};
bool Orientation(const Path &poly);
double Area(const Path &poly);
int PointInPolygon(const IntPoint &pt, const Path &path);
void SimplifyPolygon(const Path &in_poly, Paths &out_polys,
PolyFillType fillType = pftEvenOdd);
void SimplifyPolygons(const Paths &in_polys, Paths &out_polys,
PolyFillType fillType = pftEvenOdd);
void SimplifyPolygons(Paths &polys, PolyFillType fillType = pftEvenOdd);
void CleanPolygon(const Path &in_poly, Path &out_poly, double distance = 1.415);
void CleanPolygon(Path &poly, double distance = 1.415);
void CleanPolygons(const Paths &in_polys, Paths &out_polys,
double distance = 1.415);
void CleanPolygons(Paths &polys, double distance = 1.415);
void MinkowskiSum(const Path &pattern, const Path &path, Paths &solution,
bool pathIsClosed);
void MinkowskiSum(const Path &pattern, const Paths &paths, Paths &solution,
bool pathIsClosed);
void MinkowskiDiff(const Path &poly1, const Path &poly2, Paths &solution);
void PolyTreeToPaths(const PolyTree &polytree, Paths &paths);
void ClosedPathsFromPolyTree(const PolyTree &polytree, Paths &paths);
void OpenPathsFromPolyTree(PolyTree &polytree, Paths &paths);
void ReversePath(Path &p);
void ReversePaths(Paths &p);
struct IntRect {
cInt left;
cInt top;
cInt right;
cInt bottom;
};
// enums that are used internally ...
enum EdgeSide { esLeft = 1, esRight = 2 };
// forward declarations (for stuff used internally) ...
struct TEdge;
struct IntersectNode;
struct LocalMinimum;
struct OutPt;
struct OutRec;
struct Join;
typedef std::vector<OutRec *> PolyOutList;
typedef std::vector<TEdge *> EdgeList;
typedef std::vector<Join *> JoinList;
typedef std::vector<IntersectNode *> IntersectList;
//------------------------------------------------------------------------------
// ClipperBase is the ancestor to the Clipper class. It should not be
// instantiated directly. This class simply abstracts the conversion of sets of
// polygon coordinates into edge objects that are stored in a LocalMinima list.
class ClipperBase {
public:
ClipperBase();
virtual ~ClipperBase();
virtual bool AddPath(const Path &pg, PolyType PolyTyp, bool Closed);
bool AddPaths(const Paths &ppg, PolyType PolyTyp, bool Closed);
virtual void Clear();
IntRect GetBounds();
bool PreserveCollinear() { return m_PreserveCollinear; };
void PreserveCollinear(bool value) { m_PreserveCollinear = value; };
protected:
void DisposeLocalMinimaList();
TEdge *AddBoundsToLML(TEdge *e, bool IsClosed);
virtual void Reset();
TEdge *ProcessBound(TEdge *E, bool IsClockwise);
void InsertScanbeam(const cInt Y);
bool PopScanbeam(cInt &Y);
bool LocalMinimaPending();
bool PopLocalMinima(cInt Y, const LocalMinimum *&locMin);
OutRec *CreateOutRec();
void DisposeAllOutRecs();
void DisposeOutRec(PolyOutList::size_type index);
void SwapPositionsInAEL(TEdge *edge1, TEdge *edge2);
void DeleteFromAEL(TEdge *e);
void UpdateEdgeIntoAEL(TEdge *&e);
typedef std::vector<LocalMinimum> MinimaList;
MinimaList::iterator m_CurrentLM;
MinimaList m_MinimaList;
bool m_UseFullRange;
EdgeList m_edges;
bool m_PreserveCollinear;
bool m_HasOpenPaths;
PolyOutList m_PolyOuts;
TEdge *m_ActiveEdges;
typedef std::priority_queue<cInt> ScanbeamList;
ScanbeamList m_Scanbeam;
};
//------------------------------------------------------------------------------
class Clipper : public virtual ClipperBase {
public:
Clipper(int initOptions = 0);
bool Execute(ClipType clipType, Paths &solution,
PolyFillType fillType = pftEvenOdd);
bool Execute(ClipType clipType, Paths &solution, PolyFillType subjFillType,
PolyFillType clipFillType);
bool Execute(ClipType clipType, PolyTree &polytree,
PolyFillType fillType = pftEvenOdd);
bool Execute(ClipType clipType, PolyTree &polytree, PolyFillType subjFillType,
PolyFillType clipFillType);
bool ReverseSolution() { return m_ReverseOutput; };
void ReverseSolution(bool value) { m_ReverseOutput = value; };
bool StrictlySimple() { return m_StrictSimple; };
void StrictlySimple(bool value) { m_StrictSimple = value; };
// set the callback function for z value filling on intersections (otherwise Z
// is 0)
#ifdef use_xyz
void ZFillFunction(ZFillCallback zFillFunc);
#endif
protected:
virtual bool ExecuteInternal();
private:
JoinList m_Joins;
JoinList m_GhostJoins;
IntersectList m_IntersectList;
ClipType m_ClipType;
typedef std::list<cInt> MaximaList;
MaximaList m_Maxima;
TEdge *m_SortedEdges;
bool m_ExecuteLocked;
PolyFillType m_ClipFillType;
PolyFillType m_SubjFillType;
bool m_ReverseOutput;
bool m_UsingPolyTree;
bool m_StrictSimple;
#ifdef use_xyz
ZFillCallback m_ZFill; // custom callback
#endif
void SetWindingCount(TEdge &edge);
bool IsEvenOddFillType(const TEdge &edge) const;
bool IsEvenOddAltFillType(const TEdge &edge) const;
void InsertLocalMinimaIntoAEL(const cInt botY);
void InsertEdgeIntoAEL(TEdge *edge, TEdge *startEdge);
void AddEdgeToSEL(TEdge *edge);
bool PopEdgeFromSEL(TEdge *&edge);
void CopyAELToSEL();
void DeleteFromSEL(TEdge *e);
void SwapPositionsInSEL(TEdge *edge1, TEdge *edge2);
bool IsContributing(const TEdge &edge) const;
bool IsTopHorz(const cInt XPos);
void DoMaxima(TEdge *e);
void ProcessHorizontals();
void ProcessHorizontal(TEdge *horzEdge);
void AddLocalMaxPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
OutPt *AddLocalMinPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
OutRec *GetOutRec(int idx);
void AppendPolygon(TEdge *e1, TEdge *e2);
void IntersectEdges(TEdge *e1, TEdge *e2, IntPoint &pt);
OutPt *AddOutPt(TEdge *e, const IntPoint &pt);
OutPt *GetLastOutPt(TEdge *e);
bool ProcessIntersections(const cInt topY);
void BuildIntersectList(const cInt topY);
void ProcessIntersectList();
void ProcessEdgesAtTopOfScanbeam(const cInt topY);
void BuildResult(Paths &polys);
void BuildResult2(PolyTree &polytree);
void SetHoleState(TEdge *e, OutRec *outrec);
void DisposeIntersectNodes();
bool FixupIntersectionOrder();
void FixupOutPolygon(OutRec &outrec);
void FixupOutPolyline(OutRec &outrec);
bool IsHole(TEdge *e);
bool FindOwnerFromSplitRecs(OutRec &outRec, OutRec *&currOrfl);
void FixHoleLinkage(OutRec &outrec);
void AddJoin(OutPt *op1, OutPt *op2, const IntPoint offPt);
void ClearJoins();
void ClearGhostJoins();
void AddGhostJoin(OutPt *op, const IntPoint offPt);
bool JoinPoints(Join *j, OutRec *outRec1, OutRec *outRec2);
void JoinCommonEdges();
void DoSimplePolygons();
void FixupFirstLefts1(OutRec *OldOutRec, OutRec *NewOutRec);
void FixupFirstLefts2(OutRec *InnerOutRec, OutRec *OuterOutRec);
void FixupFirstLefts3(OutRec *OldOutRec, OutRec *NewOutRec);
#ifdef use_xyz
void SetZ(IntPoint &pt, TEdge &e1, TEdge &e2);
#endif
};
//------------------------------------------------------------------------------
class ClipperOffset {
public:
ClipperOffset(double miterLimit = 2.0, double roundPrecision = 0.25);
~ClipperOffset();
void AddPath(const Path &path, JoinType joinType, EndType endType);
void AddPaths(const Paths &paths, JoinType joinType, EndType endType);
void Execute(Paths &solution, double delta);
void Execute(PolyTree &solution, double delta);
void Clear();
double MiterLimit;
double ArcTolerance;
private:
Paths m_destPolys;
Path m_srcPoly;
Path m_destPoly;
std::vector<DoublePoint> m_normals;
double m_delta, m_sinA, m_sin, m_cos;
double m_miterLim, m_StepsPerRad;
IntPoint m_lowest;
PolyNode m_polyNodes;
void FixOrientations();
void DoOffset(double delta);
void OffsetPoint(int j, int &k, JoinType jointype);
void DoSquare(int j, int k);
void DoMiter(int j, int k, double r);
void DoRound(int j, int k);
};
//------------------------------------------------------------------------------
class clipperException : public std::exception {
public:
clipperException(const char *description) : m_descr(description) {}
virtual ~clipperException() throw() {}
virtual const char *what() const throw() { return m_descr.c_str(); }
private:
std::string m_descr;
};
//------------------------------------------------------------------------------
} // ClipperLib namespace
#endif // clipper_hpp
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#pragma once
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include "paddle_api.h"
#include "paddle_inference_api.h"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include <include/postprocess_op.h>
#include <include/preprocess_op.h>
namespace PaddleOCR {
class DBDetector {
public:
explicit DBDetector(const std::string &model_dir, bool use_gpu = false,
const int gpu_id = 0, const int max_side_len = 960) {
LoadModel(model_dir, use_gpu);
this->max_side_len_ = max_side_len;
}
// Load Paddle inference model
void LoadModel(const std::string &model_dir, bool use_gpu,
const int min_subgraph_size = 3, const int batch_size = 1,
const int gpu_id = 0);
// Run predictor
void Run(cv::Mat &img, std::vector<std::vector<std::vector<int>>> &boxes);
private:
std::shared_ptr<PaddlePredictor> predictor_;
int max_side_len_ = 960;
std::vector<float> mean_ = {0.485f, 0.456f, 0.406f};
std::vector<float> scale_ = {1 / 0.229f, 1 / 0.224f, 1 / 0.225f};
bool is_scale_ = true;
// pre-process
ResizeImgType0 resize_op_;
Normalize normalize_op_;
Permute permute_op_;
// post-process
PostProcessor post_processor_;
};
} // namespace PaddleOCR
\ No newline at end of file
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include "paddle_api.h"
#include "paddle_inference_api.h"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include <include/postprocess_op.h>
#include <include/preprocess_op.h>
namespace PaddleOCR {
class CRNNRecognizer {
public:
explicit CRNNRecognizer(const std::string &model_dir,
const string label_path = "./tools/ppocr_keys_v1.txt",
bool use_gpu = false, const int gpu_id = 0) {
LoadModel(model_dir, use_gpu);
this->label_list_ = ReadDict(label_path);
}
// Load Paddle inference model
void LoadModel(const std::string &model_dir, bool use_gpu,
const int gpu_id = 0, const int min_subgraph_size = 3,
const int batch_size = 1);
void Run(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat &img);
private:
std::shared_ptr<PaddlePredictor> predictor_;
std::vector<std::string> label_list_;
std::vector<float> mean_ = {0.5f, 0.5f, 0.5f};
std::vector<float> scale_ = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
bool is_scale_ = true;
// pre-process
CrnnResizeImg resize_op_;
Normalize normalize_op_;
Permute permute_op_;
// post-process
PostProcessor post_processor_;
cv::Mat get_rotate_crop_image(const cv::Mat &srcimage,
std::vector<std::vector<int>> box);
std::vector<std::string> ReadDict(const std::string &path);
template <class ForwardIterator>
inline size_t argmax(ForwardIterator first, ForwardIterator last) {
return std::distance(first, std::max_element(first, last));
}
}; // class CrnnRecognizer
} // namespace PaddleOCR
\ No newline at end of file
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#pragma once
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include "include/clipper.h"
using namespace std;
namespace PaddleOCR {
inline std::vector<std::string> ReadDict(std::string path) {
std::ifstream in(path);
std::string filename;
std::string line;
std::vector<std::string> m_vec;
if (in) {
while (getline(in, line)) {
m_vec.push_back(line);
}
} else {
std::cout << "no such file" << std::endl;
}
return m_vec;
}
template <class ForwardIterator>
inline size_t Argmax(ForwardIterator first, ForwardIterator last) {
return std::distance(first, std::max_element(first, last));
}
class PostProcessor {
public:
void GetContourArea(float **box, float unclip_ratio, float &distance);
cv::RotatedRect unclip(float **box);
float **Mat2Vec(cv::Mat mat);
void quickSort_vector(std::vector<std::vector<int>> &box, int l, int r,
int axis);
std::vector<std::vector<int>>
order_points_clockwise(std::vector<std::vector<int>> pts);
float **get_mini_boxes(cv::RotatedRect box, float &ssid);
float box_score_fast(float **box_array, cv::Mat pred);
std::vector<std::vector<std::vector<int>>>
boxes_from_bitmap(const cv::Mat pred, const cv::Mat bitmap);
std::vector<std::vector<std::vector<int>>>
filter_tag_det_res(std::vector<std::vector<std::vector<int>>> boxes,
float ratio_h, float ratio_w, cv::Mat srcimg);
template <class ForwardIterator>
inline size_t argmax(ForwardIterator first, ForwardIterator last) {
return std::distance(first, std::max_element(first, last));
}
// CRNN
private:
void quickSort(float **s, int l, int r);
inline int _max(int a, int b) { return a >= b ? a : b; }
inline int _min(int a, int b) { return a >= b ? b : a; }
template <class T> inline T clamp(T x, T min, T max) {
if (x > max)
return max;
if (x < min)
return min;
return x;
}
inline float clampf(float x, float min, float max) {
if (x > max)
return max;
if (x < min)
return min;
return x;
}
};
} // namespace PaddleOCR
\ No newline at end of file
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#pragma once
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
using namespace std;
using namespace paddle;
namespace PaddleOCR {
class Normalize {
public:
virtual void Run(cv::Mat *im, const std::vector<float> &mean,
const std::vector<float> &scale, const bool is_scale = true);
};
// RGB -> CHW
class Permute {
public:
virtual void Run(const cv::Mat *im, float *data);
};
// RGB -> CHW
class ResizeImgType0 {
public:
virtual void Run(const cv::Mat &img, cv::Mat &resize_img, int max_size_len,
float &ratio_h, float &ratio_w);
};
class CrnnResizeImg {
public:
virtual void Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
const std::vector<int> rec_image_shape = {3, 32, 320});
};
} // namespace PaddleOCR
\ No newline at end of file
此差异已折叠。
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include <include/ocr_det.h>
#include <include/ocr_rec.h>
using namespace std;
using namespace cv;
using namespace PaddleOCR;
int main(int argc, char **argv) {
if (argc < 4) {
std::cerr << "[ERROR] usage: " << argv[0]
<< " det_model_file rec_model_file image_path\n";
exit(1);
}
std::string det_model_file = argv[1];
std::string rec_model_file = argv[2];
std::string img_path = argv[3];
auto start = std::chrono::system_clock::now();
cv::Mat srcimg = cv::imread(img_path, cv::IMREAD_COLOR);
DBDetector det(det_model_file);
CRNNRecognizer rec(rec_model_file);
std::vector<std::vector<std::vector<int>>> boxes;
det.Run(srcimg, boxes);
rec.Run(boxes, srcimg);
auto end = std::chrono::system_clock::now();
auto duration =
std::chrono::duration_cast<std::chrono::microseconds>(end - start);
std::cout << "花费了"
<< double(duration.count()) *
std::chrono::microseconds::period::num /
std::chrono::microseconds::period::den
<< "秒" << std::endl;
return 0;
}
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include "paddle_api.h"
#include "paddle_inference_api.h"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include <include/ocr_det.h>
namespace PaddleOCR {
void DBDetector::LoadModel(const std::string &model_dir, bool use_gpu,
const int gpu_id, const int min_subgraph_size,
const int batch_size) {
AnalysisConfig config;
config.SetModel(model_dir + "/model", model_dir + "/params");
// for cpu
config.DisableGpu();
config.EnableMKLDNN(); // 开启MKLDNN加速
config.SetCpuMathLibraryNumThreads(10);
// 使用ZeroCopyTensor,此处必须设置为false
config.SwitchUseFeedFetchOps(false);
// 若输入为多个,此处必须设置为true
config.SwitchSpecifyInputNames(true);
// config.SwitchIrDebug(true); //
// 可视化调试选项,若开启,则会在每个图优化过程后生成dot文件
// config.SwitchIrOptim(false);// 默认为true。如果设置为false,关闭所有优化
config.EnableMemoryOptim(); // 开启内存/显存复用
this->predictor_ = CreatePaddlePredictor(config);
// predictor_ = std::move(CreatePaddlePredictor(config)); // PaddleDetection
// usage
}
void DBDetector::Run(cv::Mat &img,
std::vector<std::vector<std::vector<int>>> &boxes) {
float ratio_h{};
float ratio_w{};
cv::Mat srcimg;
cv::Mat resize_img;
img.copyTo(srcimg);
this->resize_op_.Run(img, resize_img, this->max_side_len_, ratio_h, ratio_w);
this->normalize_op_.Run(&resize_img, this->mean_, this->scale_,
this->is_scale_);
float *input = new float[1 * 3 * resize_img.rows * resize_img.cols];
this->permute_op_.Run(&resize_img, input);
auto input_names = this->predictor_->GetInputNames();
auto input_t = this->predictor_->GetInputTensor(input_names[0]);
input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
input_t->copy_from_cpu(input);
this->predictor_->ZeroCopyRun();
std::vector<float> out_data;
auto output_names = this->predictor_->GetOutputNames();
auto output_t = this->predictor_->GetOutputTensor(output_names[0]);
std::vector<int> output_shape = output_t->shape();
int out_num = std::accumulate(output_shape.begin(), output_shape.end(), 1,
std::multiplies<int>());
out_data.resize(out_num);
output_t->copy_to_cpu(out_data.data());
int n2 = output_shape[2];
int n3 = output_shape[3];
int n = n2 * n3;
float *pred = new float[n];
unsigned char *cbuf = new unsigned char[n];
for (int i = 0; i < n; i++) {
pred[i] = float(out_data[i]);
cbuf[i] = (unsigned char)((out_data[i]) * 255);
}
cv::Mat cbuf_map(n2, n3, CV_8UC1, (unsigned char *)cbuf);
cv::Mat pred_map(n2, n3, CV_32F, (float *)pred);
const double threshold = 0.3 * 255;
const double maxvalue = 255;
cv::Mat bit_map;
cv::threshold(cbuf_map, bit_map, threshold, maxvalue, cv::THRESH_BINARY);
boxes = post_processor_.boxes_from_bitmap(pred_map, bit_map);
boxes = post_processor_.filter_tag_det_res(boxes, ratio_h, ratio_w, srcimg);
//// visualization
cv::Point rook_points[boxes.size()][4];
for (int n = 0; n < boxes.size(); n++) {
for (int m = 0; m < boxes[0].size(); m++) {
rook_points[n][m] = cv::Point(int(boxes[n][m][0]), int(boxes[n][m][1]));
}
}
cv::Mat img_vis;
srcimg.copyTo(img_vis);
for (int n = 0; n < boxes.size(); n++) {
const cv::Point *ppt[1] = {rook_points[n]};
int npt[] = {4};
cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
}
imwrite("./det_res.png", img_vis);
std::cout << "The detection visualized image saved in ./det_res.png"
<< std::endl;
delete[] input;
delete[] pred;
delete[] cbuf;
}
} // namespace PaddleOCR
\ No newline at end of file
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include "paddle_api.h"
#include "paddle_inference_api.h"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include <include/ocr_rec.h>
namespace PaddleOCR {
void CRNNRecognizer::Run(std::vector<std::vector<std::vector<int>>> boxes,
cv::Mat &img) {
cv::Mat srcimg;
img.copyTo(srcimg);
cv::Mat crop_img;
cv::Mat resize_img;
std::cout << "The predicted text is :" << std::endl;
int index = 0;
for (int i = boxes.size() - 1; i >= 0; i--) {
crop_img = get_rotate_crop_image(srcimg, boxes[i]);
float wh_ratio = float(crop_img.cols) / float(crop_img.rows);
this->resize_op_.Run(crop_img, resize_img, wh_ratio);
this->normalize_op_.Run(&resize_img, this->mean_, this->scale_,
this->is_scale_);
float *input = new float[1 * 3 * resize_img.rows * resize_img.cols];
this->permute_op_.Run(&resize_img, input);
auto input_names = this->predictor_->GetInputNames();
auto input_t = this->predictor_->GetInputTensor(input_names[0]);
input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
input_t->copy_from_cpu(input);
this->predictor_->ZeroCopyRun();
std::vector<int64_t> rec_idx;
auto output_names = this->predictor_->GetOutputNames();
auto output_t = this->predictor_->GetOutputTensor(output_names[0]);
auto rec_idx_lod = output_t->lod();
auto shape_out = output_t->shape();
int out_num = std::accumulate(shape_out.begin(), shape_out.end(), 1,
std::multiplies<int>());
rec_idx.resize(out_num);
output_t->copy_to_cpu(rec_idx.data());
std::vector<int> pred_idx;
for (int n = int(rec_idx_lod[0][0]); n < int(rec_idx_lod[0][1]); n++) {
pred_idx.push_back(int(rec_idx[n]));
}
if (pred_idx.size() < 1e-3)
continue;
index += 1;
std::cout << index << "\t";
for (int n = 0; n < pred_idx.size(); n++) {
std::cout << label_list_[pred_idx[n]];
}
std::vector<float> predict_batch;
auto output_t_1 = this->predictor_->GetOutputTensor(output_names[1]);
auto predict_lod = output_t_1->lod();
auto predict_shape = output_t_1->shape();
int out_num_1 = std::accumulate(predict_shape.begin(), predict_shape.end(),
1, std::multiplies<int>());
predict_batch.resize(out_num_1);
output_t_1->copy_to_cpu(predict_batch.data());
int argmax_idx;
int blank = predict_shape[1];
float score = 0.f;
int count = 0;
float max_value = 0.0f;
for (int n = predict_lod[0][0]; n < predict_lod[0][1] - 1; n++) {
argmax_idx = int(argmax(&predict_batch[n * predict_shape[1]],
&predict_batch[(n + 1) * predict_shape[1]]));
max_value =
float(*std::max_element(&predict_batch[n * predict_shape[1]],
&predict_batch[(n + 1) * predict_shape[1]]));
if (blank - 1 - argmax_idx > 1e-5) {
score += max_value;
count += 1;
}
}
score /= count;
std::cout << "\tscore: " << score << std::endl;
delete[] input;
}
}
void CRNNRecognizer::LoadModel(const std::string &model_dir, bool use_gpu,
const int gpu_id, const int min_subgraph_size,
const int batch_size) {
AnalysisConfig config;
config.SetModel(model_dir + "/model", model_dir + "/params");
// for cpu
config.DisableGpu();
config.EnableMKLDNN(); // 开启MKLDNN加速
config.SetCpuMathLibraryNumThreads(10);
// 使用ZeroCopyTensor,此处必须设置为false
config.SwitchUseFeedFetchOps(false);
// 若输入为多个,此处必须设置为true
config.SwitchSpecifyInputNames(true);
// config.SwitchIrDebug(true); //
// 可视化调试选项,若开启,则会在每个图优化过程后生成dot文件
// config.SwitchIrOptim(false);// 默认为true。如果设置为false,关闭所有优化
config.EnableMemoryOptim(); // 开启内存/显存复用
this->predictor_ = CreatePaddlePredictor(config);
}
cv::Mat
CRNNRecognizer::get_rotate_crop_image(const cv::Mat &srcimage,
std::vector<std::vector<int>> box) {
cv::Mat image;
srcimage.copyTo(image);
std::vector<std::vector<int>> points = box;
int x_collect[4] = {box[0][0], box[1][0], box[2][0], box[3][0]};
int y_collect[4] = {box[0][1], box[1][1], box[2][1], box[3][1]};
int left = int(*std::min_element(x_collect, x_collect + 4));
int right = int(*std::max_element(x_collect, x_collect + 4));
int top = int(*std::min_element(y_collect, y_collect + 4));
int bottom = int(*std::max_element(y_collect, y_collect + 4));
cv::Mat img_crop;
image(cv::Rect(left, top, right - left, bottom - top)).copyTo(img_crop);
for (int i = 0; i < points.size(); i++) {
points[i][0] -= left;
points[i][1] -= top;
}
int img_crop_width = int(sqrt(pow(points[0][0] - points[1][0], 2) +
pow(points[0][1] - points[1][1], 2)));
int img_crop_height = int(sqrt(pow(points[0][0] - points[3][0], 2) +
pow(points[0][1] - points[3][1], 2)));
cv::Point2f pts_std[4];
pts_std[0] = cv::Point2f(0., 0.);
pts_std[1] = cv::Point2f(img_crop_width, 0.);
pts_std[2] = cv::Point2f(img_crop_width, img_crop_height);
pts_std[3] = cv::Point2f(0.f, img_crop_height);
cv::Point2f pointsf[4];
pointsf[0] = cv::Point2f(points[0][0], points[0][1]);
pointsf[1] = cv::Point2f(points[1][0], points[1][1]);
pointsf[2] = cv::Point2f(points[2][0], points[2][1]);
pointsf[3] = cv::Point2f(points[3][0], points[3][1]);
cv::Mat M = cv::getPerspectiveTransform(pointsf, pts_std);
cv::Mat dst_img;
cv::warpPerspective(img_crop, dst_img, M,
cv::Size(img_crop_width, img_crop_height),
cv::BORDER_REPLICATE);
if (float(dst_img.rows) >= float(dst_img.cols) * 1.5) {
cv::Mat srcCopy = cv::Mat(dst_img.rows, dst_img.cols, dst_img.depth());
cv::transpose(dst_img, srcCopy);
cv::flip(srcCopy, srcCopy, 0);
return srcCopy;
} else {
return dst_img;
}
}
std::vector<std::string> CRNNRecognizer::ReadDict(const std::string &path) {
std::ifstream in(path);
std::string filename;
std::string line;
std::vector<std::string> m_vec;
if (in) {
while (getline(in, line)) {
m_vec.push_back(line);
}
} else {
std::cout << "no such file" << std::endl;
}
return m_vec;
}
} // namespace PaddleOCR
\ No newline at end of file
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include <include/postprocess_op.h>
namespace PaddleOCR {
void PostProcessor::GetContourArea(float **box, float unclip_ratio,
float &distance) {
int pts_num = 4;
float area = 0.0f;
float dist = 0.0f;
for (int i = 0; i < pts_num; i++) {
area += box[i][0] * box[(i + 1) % pts_num][1] -
box[i][1] * box[(i + 1) % pts_num][0];
dist += sqrtf((box[i][0] - box[(i + 1) % pts_num][0]) *
(box[i][0] - box[(i + 1) % pts_num][0]) +
(box[i][1] - box[(i + 1) % pts_num][1]) *
(box[i][1] - box[(i + 1) % pts_num][1]));
}
area = fabs(float(area / 2.0));
distance = area * unclip_ratio / dist;
}
cv::RotatedRect PostProcessor::unclip(float **box) {
float unclip_ratio = 2.0;
float distance = 1.0;
GetContourArea(box, unclip_ratio, distance);
ClipperLib::ClipperOffset offset;
ClipperLib::Path p;
p << ClipperLib::IntPoint(int(box[0][0]), int(box[0][1]))
<< ClipperLib::IntPoint(int(box[1][0]), int(box[1][1]))
<< ClipperLib::IntPoint(int(box[2][0]), int(box[2][1]))
<< ClipperLib::IntPoint(int(box[3][0]), int(box[3][1]));
offset.AddPath(p, ClipperLib::jtRound, ClipperLib::etClosedPolygon);
ClipperLib::Paths soln;
offset.Execute(soln, distance);
std::vector<cv::Point2f> points;
for (int j = 0; j < soln.size(); j++) {
for (int i = 0; i < soln[soln.size() - 1].size(); i++) {
points.emplace_back(soln[j][i].X, soln[j][i].Y);
}
}
cv::RotatedRect res = cv::minAreaRect(points);
return res;
}
float **PostProcessor::Mat2Vec(cv::Mat mat) {
auto **array = new float *[mat.rows];
for (int i = 0; i < mat.rows; ++i)
array[i] = new float[mat.cols];
for (int i = 0; i < mat.rows; ++i) {
for (int j = 0; j < mat.cols; ++j) {
array[i][j] = mat.at<float>(i, j);
}
}
return array;
}
void PostProcessor::quickSort(float **s, int l, int r) {
if (l < r) {
int i = l, j = r;
float x = s[l][0];
float *xp = s[l];
while (i < j) {
while (i < j && s[j][0] >= x)
j--;
if (i < j)
std::swap(s[i++], s[j]);
while (i < j && s[i][0] < x)
i++;
if (i < j)
std::swap(s[j--], s[i]);
}
s[i] = xp;
quickSort(s, l, i - 1);
quickSort(s, i + 1, r);
}
}
void PostProcessor::quickSort_vector(std::vector<std::vector<int>> &box, int l,
int r, int axis) {
if (l < r) {
int i = l, j = r;
int x = box[l][axis];
std::vector<int> xp(box[l]);
while (i < j) {
while (i < j && box[j][axis] >= x)
j--;
if (i < j)
std::swap(box[i++], box[j]);
while (i < j && box[i][axis] < x)
i++;
if (i < j)
std::swap(box[j--], box[i]);
}
box[i] = xp;
quickSort_vector(box, l, i - 1, axis);
quickSort_vector(box, i + 1, r, axis);
}
}
std::vector<std::vector<int>>
PostProcessor::order_points_clockwise(std::vector<std::vector<int>> pts) {
std::vector<std::vector<int>> box = pts;
quickSort_vector(box, 0, int(box.size() - 1), 0);
std::vector<std::vector<int>> leftmost = {box[0], box[1]};
std::vector<std::vector<int>> rightmost = {box[2], box[3]};
if (leftmost[0][1] > leftmost[1][1])
std::swap(leftmost[0], leftmost[1]);
if (rightmost[0][1] > rightmost[1][1])
std::swap(rightmost[0], rightmost[1]);
std::vector<std::vector<int>> rect = {leftmost[0], rightmost[0], rightmost[1],
leftmost[1]};
return rect;
}
float **PostProcessor::get_mini_boxes(cv::RotatedRect box, float &ssid) {
ssid = box.size.width >= box.size.height ? box.size.height : box.size.width;
cv::Mat points;
cv::boxPoints(box, points);
// sorted box points
auto array = Mat2Vec(points);
quickSort(array, 0, 3);
float *idx1 = array[0], *idx2 = array[1], *idx3 = array[2], *idx4 = array[3];
if (array[3][1] <= array[2][1]) {
idx2 = array[3];
idx3 = array[2];
} else {
idx2 = array[2];
idx3 = array[3];
}
if (array[1][1] <= array[0][1]) {
idx1 = array[1];
idx4 = array[0];
} else {
idx1 = array[0];
idx4 = array[1];
}
array[0] = idx1;
array[1] = idx2;
array[2] = idx3;
array[3] = idx4;
return array;
}
float PostProcessor::box_score_fast(float **box_array, cv::Mat pred) {
auto array = box_array;
int width = pred.cols;
int height = pred.rows;
float box_x[4] = {array[0][0], array[1][0], array[2][0], array[3][0]};
float box_y[4] = {array[0][1], array[1][1], array[2][1], array[3][1]};
int xmin = clamp(int(std::floor(*(std::min_element(box_x, box_x + 4)))), 0,
width - 1);
int xmax = clamp(int(std::ceil(*(std::max_element(box_x, box_x + 4)))), 0,
width - 1);
int ymin = clamp(int(std::floor(*(std::min_element(box_y, box_y + 4)))), 0,
height - 1);
int ymax = clamp(int(std::ceil(*(std::max_element(box_y, box_y + 4)))), 0,
height - 1);
cv::Mat mask;
mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
cv::Point root_point[4];
root_point[0] = cv::Point(int(array[0][0]) - xmin, int(array[0][1]) - ymin);
root_point[1] = cv::Point(int(array[1][0]) - xmin, int(array[1][1]) - ymin);
root_point[2] = cv::Point(int(array[2][0]) - xmin, int(array[2][1]) - ymin);
root_point[3] = cv::Point(int(array[3][0]) - xmin, int(array[3][1]) - ymin);
const cv::Point *ppt[1] = {root_point};
int npt[] = {4};
cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
cv::Mat croppedImg;
pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1))
.copyTo(croppedImg);
auto score = cv::mean(croppedImg, mask)[0];
return score;
}
std::vector<std::vector<std::vector<int>>>
PostProcessor::boxes_from_bitmap(const cv::Mat pred, const cv::Mat bitmap) {
const int min_size = 3;
const int max_candidates = 1000;
const float box_thresh = 0.5;
int width = bitmap.cols;
int height = bitmap.rows;
std::vector<std::vector<cv::Point>> contours;
std::vector<cv::Vec4i> hierarchy;
cv::findContours(bitmap, contours, hierarchy, cv::RETR_LIST,
cv::CHAIN_APPROX_SIMPLE);
int num_contours =
contours.size() >= max_candidates ? max_candidates : contours.size();
std::vector<std::vector<std::vector<int>>> boxes;
for (int _i = 0; _i < num_contours; _i++) {
float ssid;
cv::RotatedRect box = cv::minAreaRect(contours[_i]);
auto array = get_mini_boxes(box, ssid);
auto box_for_unclip = array;
// end get_mini_box
if (ssid < min_size) {
continue;
}
float score;
score = box_score_fast(array, pred);
// end box_score_fast
if (score < box_thresh)
continue;
// start for unclip
cv::RotatedRect points = unclip(box_for_unclip);
// end for unclip
cv::RotatedRect clipbox = points;
auto cliparray = get_mini_boxes(clipbox, ssid);
if (ssid < min_size + 2)
continue;
int dest_width = pred.cols;
int dest_height = pred.rows;
std::vector<std::vector<int>> intcliparray;
for (int num_pt = 0; num_pt < 4; num_pt++) {
std::vector<int> a{int(clampf(roundf(cliparray[num_pt][0] / float(width) *
float(dest_width)),
0, float(dest_width))),
int(clampf(roundf(cliparray[num_pt][1] /
float(height) * float(dest_height)),
0, float(dest_height)))};
intcliparray.push_back(a);
}
boxes.push_back(intcliparray);
} // end for
return boxes;
}
std::vector<std::vector<std::vector<int>>> PostProcessor::filter_tag_det_res(
std::vector<std::vector<std::vector<int>>> boxes, float ratio_h,
float ratio_w, cv::Mat srcimg) {
int oriimg_h = srcimg.rows;
int oriimg_w = srcimg.cols;
std::vector<std::vector<std::vector<int>>> root_points;
for (int n = 0; n < boxes.size(); n++) {
boxes[n] = order_points_clockwise(boxes[n]);
for (int m = 0; m < boxes[0].size(); m++) {
boxes[n][m][0] /= ratio_w;
boxes[n][m][1] /= ratio_h;
boxes[n][m][0] = int(_min(_max(boxes[n][m][0], 0), oriimg_w - 1));
boxes[n][m][1] = int(_min(_max(boxes[n][m][1], 0), oriimg_h - 1));
}
}
for (int n = 0; n < boxes.size(); n++) {
int rect_width, rect_height;
rect_width = int(sqrt(pow(boxes[n][0][0] - boxes[n][1][0], 2) +
pow(boxes[n][0][1] - boxes[n][1][1], 2)));
rect_height = int(sqrt(pow(boxes[n][0][0] - boxes[n][3][0], 2) +
pow(boxes[n][0][1] - boxes[n][3][1], 2)));
if (rect_width <= 10 || rect_height <= 10)
continue;
root_points.push_back(boxes[n]);
}
return root_points;
}
} // namespace PaddleOCR
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include "paddle_api.h"
#include "paddle_inference_api.h"
#include <chrono>
#include <iomanip>
#include <iostream>
#include <ostream>
#include <vector>
#include <cstring>
#include <fstream>
#include <numeric>
#include <include/preprocess_op.h>
namespace PaddleOCR {
void Permute::Run(const cv::Mat *im, float *data) {
int rh = im->rows;
int rw = im->cols;
int rc = im->channels();
for (int i = 0; i < rc; ++i) {
cv::extractChannel(*im, cv::Mat(rh, rw, CV_32FC1, data + i * rh * rw), i);
}
}
void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
const std::vector<float> &scale, const bool is_scale) {
double e = 1.0;
if (is_scale) {
e /= 255.0;
}
(*im).convertTo(*im, CV_32FC3, e);
for (int h = 0; h < im->rows; h++) {
for (int w = 0; w < im->cols; w++) {
im->at<cv::Vec3f>(h, w)[0] =
(im->at<cv::Vec3f>(h, w)[0] - mean[0]) * scale[0];
im->at<cv::Vec3f>(h, w)[1] =
(im->at<cv::Vec3f>(h, w)[1] - mean[1]) * scale[1];
im->at<cv::Vec3f>(h, w)[2] =
(im->at<cv::Vec3f>(h, w)[2] - mean[2]) * scale[2];
}
}
}
void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
int max_size_len, float &ratio_h, float &ratio_w) {
int w = img.cols;
int h = img.rows;
float ratio = 1.f;
int max_wh = w >= h ? w : h;
if (max_wh > max_size_len) {
if (h > w) {
ratio = float(max_size_len) / float(h);
} else {
ratio = float(max_size_len) / float(w);
}
}
int resize_h = int(float(h) * ratio);
int resize_w = int(float(w) * ratio);
if (resize_h % 32 == 0)
resize_h = resize_h;
else if (resize_h / 32 < 1 + 1e-5)
resize_h = 32;
else
resize_h = (resize_h / 32 - 1) * 32;
if (resize_w % 32 == 0)
resize_w = resize_w;
else if (resize_w / 32 < 1)
resize_w = 32;
else
resize_w = (resize_w / 32 - 1) * 32;
cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
ratio_h = float(resize_h) / float(h);
ratio_w = float(resize_w) / float(w);
}
void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
const std::vector<int> rec_image_shape) {
int imgC, imgH, imgW;
imgC = rec_image_shape[0];
imgH = rec_image_shape[1];
imgW = rec_image_shape[2];
imgW = int(32 * wh_ratio);
float ratio = float(img.cols) / float(img.rows);
int resize_w, resize_h;
if (ceilf(imgH * ratio) > imgW)
resize_w = imgW;
else
resize_w = int(ceilf(imgH * ratio));
cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
cv::INTER_LINEAR);
}
} // namespace PaddleOCR
\ No newline at end of file
LIB_DIR=/paddle/code/gry/Paddle/build/fluid_inference_install_dir/
CUDA_LIB_DIR=/usr/local/cuda/lib64
CUDNN_LIB_DIR=/usr/lib/x86_64-linux-gnu/
TENSORRT_ROOT_DIR=YOUR_TENSORRT_ROOT_DIR
BUILD_DIR=build
rm -rf ${BUILD_DIR}
mkdir ${BUILD_DIR}
cd ${BUILD_DIR}
cmake .. \
-DPADDLE_LIB=${LIB_DIR} \
-DWITH_MKL=ON \
-DDEMO_NAME=ocr_system \
-DWITH_GPU=OFF \
-DWITH_STATIC_LIB=OFF \
-DUSE_TENSORRT=OFF \
-DCUDNN_LIB=${CUDNN_LIB_DIR} \
-DCUDA_LIB=${CUDA_LIB_DIR} \
-DTENSORRT_ROOT=YOUR_TENSORRT_ROOT_DIR
make -j
此差异已折叠。
./build/ocr_system ./inference/det_db/ ./inference/rec_crnn/ ../../doc/imgs/11.jpg
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册