Load jpeg from memory

This commit is contained in:
Kagami Hiiragi
2018-04-05 13:05:30 +03:00
parent c627f1a583
commit 44e80c999f
8 changed files with 119 additions and 35 deletions

View File

@@ -2,5 +2,5 @@ package dlib
// #cgo pkg-config: dlib-1 // #cgo pkg-config: dlib-1
// #cgo CFLAGS: -Wall -O3 -march=native // #cgo CFLAGS: -Wall -O3 -march=native
// #cgo LDFLAGS: -lgif // #cgo LDFLAGS: -ljpeg
import "C" import "C"

View File

@@ -16,12 +16,6 @@ func (e SerializationError) Error() string {
return string(e) return string(e)
} }
type RecognizeError string
func (e RecognizeError) Error() string {
return string(e)
}
type UnknownError string type UnknownError string
func (e UnknownError) Error() string { func (e UnknownError) Error() string {
@@ -35,8 +29,6 @@ func makeError(s string, code int) error {
return ImageLoadError(s) return ImageLoadError(s)
case C.SERIALIZATION_ERROR: case C.SERIALIZATION_ERROR:
return SerializationError(s) return SerializationError(s)
case C.RECOGNIZE_ERROR:
return RecognizeError(s)
default: default:
return UnknownError(s) return UnknownError(s)
} }

View File

@@ -1,8 +1,8 @@
#include <dlib/dnn.h> #include <dlib/dnn.h>
#include <dlib/image_io.h> #include <dlib/image_loader/image_loader.h>
#include <dlib/image_processing/frontal_face_detector.h> #include <dlib/image_processing/frontal_face_detector.h>
#include "facerec.h" #include "facerec.h"
#include "jpeg_mem_loader.h"
using namespace dlib; using namespace dlib;
@@ -36,6 +36,7 @@ using anet_type = loss_metric<fc_no_bias<128,avg_pool_everything<
typedef matrix<float,0,1> descriptor; typedef matrix<float,0,1> descriptor;
typedef std::pair<std::vector<rectangle>, std::vector<descriptor>> faces; typedef std::pair<std::vector<rectangle>, std::vector<descriptor>> faces;
static const size_t RECT_SIZE = 4 * sizeof(long); static const size_t RECT_SIZE = 4 * sizeof(long);
static const size_t DESCR_SIZE = 128 * sizeof(float); static const size_t DESCR_SIZE = 128 * sizeof(float);
@@ -53,10 +54,7 @@ public:
} }
// TODO(Kagami): Jittering? // TODO(Kagami): Jittering?
faces Recognize(const char* img_path, int max_faces) { faces Recognize(matrix<rgb_pixel>& img, int max_faces) {
matrix<rgb_pixel> img;
load_image(img, img_path);
std::vector<rectangle> rects = detector_(img); std::vector<rectangle> rects = detector_(img);
std::vector<descriptor> descrs; std::vector<descriptor> descrs;
@@ -98,11 +96,14 @@ facerec* facerec_init(const char* model_dir) {
return rec; return rec;
} }
faceret* facerec_recognize(facerec* rec, const char* img_path, int max_faces) { faceret* facerec_recognize(facerec* rec, const uint8_t* img_data, int len, int max_faces) {
faceret* ret = (faceret*)calloc(1, sizeof(faceret)); faceret* ret = (faceret*)calloc(1, sizeof(faceret));
FaceRec* cls = (FaceRec*)(rec->cls); FaceRec* cls = (FaceRec*)(rec->cls);
matrix<rgb_pixel> img;
try { try {
faces faces = cls->Recognize(img_path, max_faces); // TODO(Kagami): Support more file types?
load_mem_jpeg(img, img_data, len);
faces faces = cls->Recognize(img, max_faces);
std::vector<rectangle> rects = faces.first; std::vector<rectangle> rects = faces.first;
std::vector<descriptor> descrs = faces.second; std::vector<descriptor> descrs = faces.second;
ret->num_faces = descrs.size(); ret->num_faces = descrs.size();

View File

@@ -6,6 +6,8 @@ package dlib
import "C" import "C"
import ( import (
"image" "image"
"io/ioutil"
"os"
"unsafe" "unsafe"
) )
@@ -41,10 +43,15 @@ func NewFaceRec(modelDir string) (rec *FaceRec, err error) {
return return
} }
func (rec *FaceRec) recognize(imgPath string, maxFaces int) (faces []Face, err error) { func (rec *FaceRec) recognize(imgData []byte, maxFaces int) (faces []Face, err error) {
cImgPath := C.CString(imgPath) if len(imgData) == 0 {
defer C.free(unsafe.Pointer(cImgPath)) err = ImageLoadError("Empty image")
ret := C.facerec_recognize(rec.p, cImgPath, C.int(maxFaces)) return
}
cImgData := (*C.uchar)(&imgData[0])
cLen := C.int(len(imgData))
cMaxFaces := C.int(maxFaces)
ret := C.facerec_recognize(rec.p, cImgData, cLen, cMaxFaces)
defer C.free(unsafe.Pointer(ret)) defer C.free(unsafe.Pointer(ret))
if ret.err_str != nil { if ret.err_str != nil {
@@ -84,20 +91,43 @@ func (rec *FaceRec) recognize(imgPath string, maxFaces int) (faces []Face, err e
return return
} }
// Recognize all image faces. // Convenient method.
// Empty list is returned if there are no faces, error is returned if func (rec *FaceRec) recognizeFile(imgPath string, maxFaces int) (face []Face, err error) {
// there was some error while decoding/processing image. fd, err := os.Open(imgPath)
func (rec *FaceRec) Recognize(imgPath string) (faces []Face, err error) {
return rec.recognize(imgPath, 0)
}
// Recognize if image has single face or return nil otherwise.
func (rec *FaceRec) RecognizeSingle(imgPath string) (face *Face, err error) {
faces, err := rec.recognize(imgPath, 1)
if err != nil { if err != nil {
return return
} }
if len(faces) != 1 { imgData, err := ioutil.ReadAll(fd)
if err != nil {
return
}
return rec.recognize(imgData, maxFaces)
}
// Recognize all image faces.
// Empty list is returned if there are no faces, error is returned if
// there was some error while decoding/processing image.
func (rec *FaceRec) Recognize(imgData []byte) (faces []Face, err error) {
return rec.recognize(imgData, 0)
}
// Recognize if image has single face or return nil otherwise.
func (rec *FaceRec) RecognizeSingle(imgData []byte) (face *Face, err error) {
faces, err := rec.recognize(imgData, 1)
if err != nil || len(faces) != 1 {
return
}
face = &faces[0]
return
}
func (rec *FaceRec) RecognizeFile(imgPath string) (faces []Face, err error) {
return rec.recognizeFile(imgPath, 0)
}
func (rec *FaceRec) RecognizeSingleFile(imgPath string) (face *Face, err error) {
faces, err := rec.recognizeFile(imgPath, 1)
if err != nil || len(faces) != 1 {
return return
} }
face = &faces[0] face = &faces[0]

View File

@@ -7,7 +7,6 @@ extern "C" {
typedef enum { typedef enum {
IMAGE_LOAD_ERROR, IMAGE_LOAD_ERROR,
SERIALIZATION_ERROR, SERIALIZATION_ERROR,
RECOGNIZE_ERROR,
UNKNOWN_ERROR, UNKNOWN_ERROR,
} err_code; } err_code;
@@ -26,7 +25,7 @@ typedef struct faceret {
} faceret; } faceret;
facerec* facerec_init(const char* model_dir); facerec* facerec_init(const char* model_dir);
faceret* facerec_recognize(facerec* rec, const char* img_path, int max_faces); faceret* facerec_recognize(facerec* rec, const uint8_t* img_data, int len, int max_faces);
void facerec_free(facerec* rec); void facerec_free(facerec* rec);
#ifdef __cplusplus #ifdef __cplusplus

View File

@@ -10,7 +10,7 @@ func TestNumFaces(t *testing.T) {
t.Fatalf("Can't init face recognizer: %v", err) t.Fatalf("Can't init face recognizer: %v", err)
} }
defer rec.Close() defer rec.Close()
faces, err := rec.Recognize("testdata/pristin.jpg") faces, err := rec.RecognizeFile("testdata/pristin.jpg")
if err != nil { if err != nil {
t.Fatalf("Can't get faces: %v", err) t.Fatalf("Can't get faces: %v", err)
} }

59
jpeg_mem_loader.cc Normal file
View File

@@ -0,0 +1,59 @@
#include <stdio.h>
#include <setjmp.h>
#include <jpeglib.h>
#include <dlib/image_loader/image_loader.h>
#include "jpeg_mem_loader.h"
// Based on dlib's JPEG loader and libjpeg example.
struct jpeg_loader_error_mgr {
jpeg_error_mgr pub; // "public" fields
jmp_buf setjmp_buffer; // For return to caller
};
static void jpeg_loader_error_exit(j_common_ptr cinfo) {
// cinfo->err really points to a jpeg_loader_error_mgr struct, so coerce pointer.
jpeg_loader_error_mgr* myerr = (jpeg_loader_error_mgr*)cinfo->err;
// Return control to the setjmp point.
longjmp(myerr->setjmp_buffer, 1);
}
static void jpeg_loader_emit_message(j_common_ptr cinfo, int msg_level) {
// Just exit on any warning because file is probably corrupted.
if (msg_level < 0)
jpeg_loader_error_exit(cinfo);
}
// From memory JPEG loader because dlib lacks one.
void load_mem_jpeg(dlib::matrix<dlib::rgb_pixel>& img, const uint8_t* img_data, int len) {
jpeg_decompress_struct cinfo;
jpeg_loader_error_mgr jerr;
cinfo.err = jpeg_std_error(&jerr.pub);
jerr.pub.error_exit = jpeg_loader_error_exit;
jerr.pub.emit_message = jpeg_loader_emit_message;
if (setjmp(jerr.setjmp_buffer)) {
char buffer[JMSG_LENGTH_MAX];
(jerr.pub.format_message)((j_common_ptr)&cinfo, buffer);
jpeg_destroy_decompress(&cinfo);
throw dlib::image_load_error(std::string("jpeg_mem_loader: decode error: ") + buffer);
}
jpeg_create_decompress(&cinfo);
jpeg_mem_src(&cinfo, img_data, len);
jpeg_read_header(&cinfo, TRUE);
jpeg_start_decompress(&cinfo);
if (cinfo.output_components != 3) {
jpeg_destroy_decompress(&cinfo);
throw dlib::image_load_error("jpeg_mem_loader: unsupported pixel size");
}
img.set_size(cinfo.output_height, cinfo.output_width);
while (cinfo.output_scanline < cinfo.output_height) {
uint8_t* buffer_array[1] = { (uint8_t*)&img(cinfo.output_scanline, 0) };
jpeg_read_scanlines(&cinfo, buffer_array, 1);
}
jpeg_finish_decompress(&cinfo);
jpeg_destroy_decompress(&cinfo);
}

3
jpeg_mem_loader.h Normal file
View File

@@ -0,0 +1,3 @@
#pragma once
void load_mem_jpeg(dlib::matrix<dlib::rgb_pixel>& img, const uint8_t* img_data, int len);