Face Recognition Algorithm

REVISION HISTORY

Revision No.
Description
 Date
100
  • Initial release
    		•  04/27/2023
    101
  • Add Face Attribute Recognition
    		•  07/04/2023
    102
  • Sync with latest api & add error code
    		•  05/20/2024
    200
  • Sync with latest api & add error code
    		•  08/01/2024

    1. Algorithm Description

    Face recognition primarily distinguishes between human faces and determines whether a captured face belongs to a whitelisted individual. The entire algorithm includes face detection, face attribute recognition, facial expression recognition, facial landmark detection, face filtering, face tracking, face alignment, feature extraction, and face comparison.

    Among these, face attributes can determine five characteristics: glasses, gender, mask, beard, and age. Facial expression recognition identifies seven expressions: neutral, happy, sad, surprised, fearful, disgusted, and angry.

    • Algorithm Accuracy

      ALGO Threshold 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
      FR TAR% 99.4 98.9 98.5 98.1 97.6 96.9 96.4 95.2 93.9 93.4 92.1 90.9 89.1 86.7 85.4 84.2
      FR FAR% 2.3 1.8 1 0.7 0.4 0.1 0 0 0 0 0 0 0 0 0 0

    • Model Introduction

      model function input_size(w*h) input_format
      fr_det_y24s.img face detection 480*288 yuvsp420_nv12
      fr_det_y36s.img face detection 640*352 yuvsp420_nv12
      fr_det_y48s.img face detection 800*480 yuvsp420_nv12
      fr_feature_as.img face feature extraction(small) 112*112 BGRA8888
      fr_feature_am.img face feature extraction(medium) 112*112 BGRA8888
      fr_feature_al.img face feature extraction(large) 112*112 BGRA8888
      fr_angle_y66.img face angle estimation 64*64 yuvsp420_nv12
      fr_cos256.img face feature compare 256*512 INT16
      far101_224y_emo.img facial expression recognition 224*224 yuvsp420_nv12
      far101_224y_emo_s45.img facial expression recognition(small) 224*224 yuvsp420_nv12
      far103_224y.img facial attribute recognition 112*112 BGRA8888
      fr_blur_as.img face blur estimation (small) 112*112 BGRA8888

    2. ​​API Workflow​

    Face recognition primarily involves two processes. The first one is ​face Registration​​, which generates a facial whitelist (face database). The second one is ​Face Recognition​​ which identifies captured faces to determine if they belong to whitelisted individuals.

    2.1. Face Registration Workflow​

    Face Registration API Workflow：ALGO_FR_CreateHandle → ALGO_FR_InitHandle → ALGO_FR_SetParams → ALGO_FR_GetInputAttr → ALGO_FR_Detect → ALGO_FR_AlignRegist → ALGO_FR_FeatureExtract. Finally save extracted features into face database.

    2.2. Face Recognition Workflow​

    Face Recognition API Workflow：ALGO_FR_CreateHandle → ALGO_FR_InitHandle → ALGO_FR_SetParams → ALGO_FR_GetInputAttr → ALGO_FR_Detect → ALGO_FR_Align → ALGO_FR_FeatureExtract → ALGO_FR_FeatureCompare.Compare the features of the captured image with those in the database (whitelist) to determines whether it belongs to the same individual.​

    2.3. Facial Atrribute Recognition Workflow​

    Facial Atrribute Recognition Workflow​：ALGO_FR_CreateHandle → ALGO_FR_InitHandle → ALGO_FR_SetParams → ALGO_FR_GetInputAttr → ALGO_FR_Detect → ALGO_FR_Attr.Facial attribute recognition is performed using captured face images.​Note: Cropped face images input to this API must be expanded with a certain margin around the face area.The expansion margins should be set to 1.25 times the width (bbox_w) and height (bbox_h) of the original bounding box.​

    2.4. Face Blur Score Workflow​

    Face Blur Score API Workflow​：ALGO_FR_CreateHandle → ALGO_FR_InitHandle → ALGO_FR_SetParams → ALGO_FR_GetInputAttr → ALGO_FR_Detect → ALGO_FR_Align → ALGO_FR_FaceBlur, send detected & aligned face area into ALGO_FR_FaceBlur to get Blur Score.

    3. Function Module API

    API Name function
    ALGO_FR_CreateHandle create handle
    ALGO_FR_InitHandle initialize handle
    ALGO_FR_GetInputAttr retrieve the model's input attributes​
    ALGO_FR_SetParams set configurable params
    ALGO_FR_Detect face detection
    ALGO_FR_Align face alignment
    ALGO_FR_Attr facial attribute recognition
    ALGO_FR_FaceQuality face angle estimation and quality filtering
    ALGO_FR_FeatureExtract feature extraction
    ALGO_FR_FeatureCompare feature comparison
    ALGO_FR_BatchFeatureCompare batch facial feature comparison(need to make one feature copy)
    ALGO_FR_BatchFeatureCompareV2 batch facial feature comparison(no need to make feature copy)
    ALGO_FR_DeinitHandle de-initialize handle
    ALGO_FR_ReleaseHandle release handle
    ALGO_FR_FaceBlur get face blur score

    3.1. ALGO_FR_CreateHandle

    • function

      create handle.

    • Syntax

      MI_S32 ALGO_FR_CreateHandle(void **handle);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.2. ALGO_FR_InitHandle

    • function

      initialize handle.

    • Syntax

      MI_S32 ALGO_FR_InitHandle(void *handle, const FrInit_t *init_info);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      init_info Algorithm initialization params, see FrInit_t for details Input

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    • Usage Examples

      {
    int ret=0;
    FrInit_t initParam;
    std::string bin = "/config/dla/ipu_firmware.bin";
    std::string detModelPath = "./models/fr_det_y24s.img";
    std::string attrModelPath = "./models/far103_224y.img";
    std::string featureModelPath = "./models/fr_feature_am.img";
    std::string cosModelPath = "./models/fr_cos256.img";
    std::string qualityModelPath = "./models/fr_angle_y66.img";
    std::string emoModelPath = "./models/far101_224y_emo.img";
    memcpy(initParam.ipu_firmware_path, bin.c_str(), bin.length()+1);
    memcpy(initParam.det_model_path, detModelPath.c_str(), detModelPath.length()+1);
    memcpy(initParam.attr_model_path, attrModelPath.c_str(), attrModelPath.length()+1);
    memcpy(initParam.feature_model_path, featureModelPath.c_str(), featureModelPath.length()+1);
    memcpy(initParam.cos_model_path, cosModelPath.c_str(), cosModelPath.length() + 1);
    memcpy(initParam.quality_model_path, qualityModelPath.c_str(), qualityModelPath.length() + 1);
    memcpy(initParam.emo_model_path, emoModelPath.c_str(), emoModelPath.length() + 1);
    initParam.create_device = true;
    initParam.destroy_device = true;
    initParam.fr_mode = FR_MODE_FACE_RECOG | FR_MODE_FACE_DET | FR_MODE_FACE_QUALITY;
    ret = ALGO_FR_InitHandle(handle, &initParam);
}

    3.3. ALGO_FR_GetInputAttr

    • Function

      Retrieve the attribute information of the face detection model, facial attribute model, and face quality model, including the input resolution and data type requirements for each model.​

    • Syntax

      MI_S32 ALGO_FR_GetInputAttr(void *handle, FrInputAttr_t *det_input_attr, FrInputAttr_t *attr_input_attr, FrInputAttr_t *quality_input_attr);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      det_input_attr pointer to the input attributes of the face detection model, see FrInputAttr_t for details Input
      attr_input_attr pointer to the input attributes of the face attribute model, see FrInputAttr_t for details Input
      quality_input_attr pointer to the input attributes of the face quality model, see FrInputAttr_t for details Input

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.4. ALGO_FR_SetParams

    • function

      ​Set configurable parameters of the algorithm.​

    • Syntax

      MI_S32 ALGO_FR_SetParams(void *handle, const FrParams_t *params);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      params algorithm parameters, seeFrParams_t for details Input

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.5. ALGO_FR_Detect

    • function

      face detection.

    • Syntax

      MI_S32 ALGO_FR_Detect(void *handle, const FrInput_t *algo_input, const FrDetectMode_e *detect_mode, FrBox_t bboxes[MAX_FR_OBJECT], MI_S32 *num_bboxes);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      algo_input input image buffer, see FrInput_t for details Input
      detect_mode detection model, see FrDetectMode_e for details Input
      bboxes array to save face detection result boxes, see FrBox_t for details Output
      num_bboxes pointer to number of detection result boxes Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.6. ALGO_FR_Align

    • function

      face alignment.

    • Syntax

      MI_S32 ALGO_FR_Align(void *handle, MI_U8* image_data, MI_S32 type, FrBox_t *detect_box, MI_U8* out_data);

    • Parameters

      Parameter Name Description Input/Output
      handle algorithm handle Input
      image_data pointer to image data buffer, note: width/height of image should be consistent of disp_width/height of FrInit_t Input
      type image format type, yuv420_nv12=1,argb8888=0 Input
      detect_box face detection boxes of input image which need to be aligned, see FrBox_t for details Input
      out_data face area image data in 112x112 ARGB8888 format, should not be null Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.6.1 ALGO_FR_AlignRegist

    • Function

      Used for the face alignment interface for registering faces.

    • Syntax

      MI_S32 ALGO_FR_AlignRegist(void *handle, MI_U8* image_data, MI_S32 width, MI_S32 height, MI_S32 type, FrBox_t *detect_box, MI_U8* out_data);

    • Parameters

      Parameter Name Description Input/Output
      handle Handle Input
      image_data Pointer to the original image data, the width and height of the image must equal the disp_width and disp_height set in the initialization parameters FrInit_t Input
      width Width of the original image data Input
      height Height of the original image data Input
      type Image type, yuv420_nv12 is 1, argb8888 is 0 Input
      detect_box Face detection box that needs to be aligned on the input image (coordinates of the box must be mapped to the original image's width and height), see FrBox_t Input
      out_data Output aligned 112x112 ARGB888 data, cannot be null Output

      Note: Coordinate transformation *width/disp_width,*height/disp_height

    • Return Value

      Return Value Description
      0 Success
      Other Failure (see Error Codes)

    • Dependencies

      Header file: sgs_fr_api.h

      Library files: libsgsalgo_fr.so, libsgsalgo_fr.a

    3.7. ALGO_FR_Attr

    • function

      face attribute recognition.

    • Syntax

      MI_S32 ALGO_FR_Attr(void *handle, const FrInput_t *algo_input, FrAttr_t *results);

    • Parameters

      Parameter Name Description Input/Output
      handle algorithm handle Input
      algo_input pointer to image data buffer, see FrInput_t for details Input
      results face attribute recognition result, see FrAttr_t for details Output

    • Description

      face image data pass in this api must be expanded with a certain margin around the face area.The expansion margins should be set to 1.25 times the width (bbox_w) and height (bbox_h) of the original bounding box.​

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.8. ALGO_FR_FaceQuality

    • function

      Face angle estimation and quality filtering.

    • Syntax

      MI_S32 ALGO_FR_FaceQuality(void *handle, const FrInput_t *algo_input, FrBox_t *detect_box, FaceQualityAngle_t *angle);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      algo_input pointer to image data buffer, see FrInput_tfor details Input
      detect_box pointer to face detection output box, see FrBox_t for details Input/Output
      angle face angle estimation result, see FaceQualityAngle_t for details Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.9. ALGO_FR_FeatureExtract

    • function

      face feature extraction.

    • Syntax

      MI_S32 ALGO_FR_FeatureExtract(void *handle, MI_U8* image_data, MI_S16* out_feature);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      image_data 112*112 image data output by ALGO_FR_Align Input
      out_feature face feature data, which is a 512-dimensional MI_S16 vector Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.10. ALGO_FR_FeatureCompare

    • function

      feature comparison.

    • Syntax

      MI_S32 ALGO_FR_FeatureCompare(MI_S16* feature1, MI_S16* feature2, MI_S32 length, MI_FLOAT* similarity);

    • Parameters

      Parameter Name Description Input/Output
      feature1 feature of first face Input
      feature2 feature of second face Input
      length feature length, equals to 512 in this version Input
      similarity similarity of two faces Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.11. ALGO_FR_BatchFeatureCompare

    • function

      batch face feature comparison(need to make one feature copy) .

    • Syntax

      MI_S32 ALGO_FR_BatchFeatureCompare(void *handle, MI_S16* single_feature, MI_S16* batch_features, MI_S32 batch, MI_FLOAT* similarity);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      single_feature destination feature(1) Input
      batch_features source features(batch) Input
      batch number of source features Input
      similarity output feature comparison similarity Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.12. ALGO_FR_BatchFeatureCompareV2

    • function

      batch face feature comparison(no need to make feature copy) .

    • Syntax

      MI_S32 ALGO_FR_BatchFeatureCompareV2(void *handle, const FrInput_t *single_feature, const FrInput_t *batch_features, MI_S32 batch, MI_FLOAT* similarity);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      single_feature source feature(1), pass in as MMA buffer Input
      batch_features destination features(batch), pass in as MMA buffer Input
      batch number of source features Input
      similarity output feature comparison similarity Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.13. ALGO_FR_DeinitHandle

    • Function

      stop algorithm and de-initialize handle.

    • Syntax

      MI_S32 ALGO_FR_DeinitHandle(void *handle);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.14. ALGO_FR_ReleaseHandle

    • function

      Release Algorithm Handle

    • Syntax

      MI_S32 ALGO_FR_ReleaseHandle(void *handle);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    3.15. ALGO_FR_FaceBlur

    • function

      get face blur score

    • Syntax

      MI_S32 ALGO_FR_FaceBlur(void *handle, const MI_U8* image_data, MI_FLOAT *blur_score);

    • Parameters

      Parameter Name Description Input/Output
      handle handle Input
      image_data 112*112 image data output by ALGO_FR_Align Input
      blur_score output pointer to blur score, which is a float range from 0.0 to 100.0, lower score means blurrier face Output

    • Return Value

      Return Value Description
      0 Success
      Others Failure(seeError Code for details)

    • Dependencies

      header files：sgs_fr_api.h

      library files：libsgsalgo_fr.so, libsgsalgo_fr.a

    4. Data Type

    Data Type Definition
    FrMode_e algorithm mode
    FrInit_t initialization algorithm parameters
    FrInputAttr_t algorithm input attributes structure
    FrParams_t configurable parameters of the algorithm
    FrInput_t model input attributes structure
    FrDetectMode_e face detection mode
    FrPoint_t face key-point coordinates structure
    FrBox_t face detection box structure
    FrAttr_t face attribute structure
    FaceQualityAngle_t face angle structure

    4.1. FrMode_e

    • Description

      Algorithm mode.

    • Definition

      typedef enum{
    FR_MODE_FACE_DET=0x01,
    FR_MODE_FACE_RECOG=0x02,
    FR_MODE_FACE_ATTR=0x04,
    FR_MODE_FACE_EMOTION=0x08,
    FR_MODE_FACE_QUALITY=0x10,
    FR_MODE_FACE_BLUR=0x20,
}FrMode_e;

    • Members

      Member Name Description
      FR_MODE_FACE_DET face detection
      FR_MODE_FACE_RECOG face recognition
      FR_MODE_FACE_ATTR facial attribute recognition
      FR_MODE_FACE_EMOTION face emotion recognition
      FR_MODE_FACE_QUALITY face angle estimation and quality filtering
      FR_MODE_FACE_BLUR face blur estimation

    • Related Data Type and APIs

      FrInit_t

      ALGO_FR_InitHandle

    4.2. FrInit_t

    • Description

      Algo initialization parameters

    • Definition

      typedef struct
{
    char ipu_firmware_path[MAX_FR_STRLEN];       // ipu_firmware.bin path
    char det_model_path[MAX_FR_STRLEN];          // detect model path, set when chose FR_MODE_FACE_DET mode
    char feature_model_path[MAX_FR_STRLEN];      // feature extraction model path, set when chose FR_MODE_FACE_RECOG mode
    char cos_model_path[MAX_FR_STRLEN];          // feature comparison model path, set when chose FR_MODE_FACE_RECOG mode
    char attr_model_path[MAX_FR_STRLEN];         // face attr model path, set when chose FR_MODE_FACE_ATTR mode
    char emo_model_path[MAX_FR_STRLEN];          // face emotion model path, set when chose FR_MODE_FACE_EMOTION mode
    char quality_model_path[MAX_FR_STRLEN];     // face angle model path, set when chose FR_MODE_FACE_QUALITY mode
    MI_BOOL create_device;                       // set false to create ipu device outside algo lib
    MI_BOOL destroy_device;                      // set false to destroy ipu device outside algo lib
    MI_U8 fr_mode;
    char blur_model_path[MAX_FR_STRLEN];         // face blur model path, set when choose
} FrInit_t;

    • Members

      Member Name Description
      ipu_firmware_path file path of ipu firmware
      det_model_path path of detection model
      feature_model_path path of feature extraction model
      cos_model_path path of similarity model
      attr_model_path path of face attribute recognition model
      emo_model_path path of face emotion recognition model
      quality_model_path path of face quality estimation model
      create_device Whether to create the IPUDevice inside the algorithm library. Default is true (created internally). Set to false if the algorithm library is called concurrently, and manually create the IPUDevice externally
      destroy_device Whether to destroy the IPUDevice inside the algorithm library. Default is true (destroyed internally). Set to false if the algorithm library is called concurrently, and manually destroy the IPUDevice externally
      fr_mode algorithm mode, default is face recognition mode
      blur_model_path path of face blur estimation model

    • Related APis

      FrMode_e

      ALGO_FR_InitHandle

    4.3 FrInputAttr_t

    • Description

      model input attribute structure

    • Definition

      typedef struct
{
    MI_U32 width;
    MI_U32 height;
    MI_IPU_ELEMENT_FORMAT format;
} FrInputAttr_t;

    • Members

      Member Name Description
      width width of model input
      height height of model input
      format pixel format of model input

    • Related Data Type and APis

      ALGO_FR_GetInputAttr

    4.4. FrParams_t

    • Description

      configurable parameters of the algorithm

    • Definition

      typedef struct
{
    MI_S32 disp_width;
    MI_S32 disp_height;

    // detect
    MI_FLOAT det_threshold;
    MI_S32  box_min_size;
    MI_FLOAT face_angle_ratio;
    MI_FLOAT eye_distance;

    // attr, emotion
    MI_FLOAT attr_threshold;

    // track
    MI_FLOAT motion_sensitive;

    // face quality
    MI_FLOAT face_quality_threshold;
} FrParams_t;

    • Members

      Member Name Description
      disp_width Width of he video stream resolution (used for mapping detection box coordinates and as the original image size for face alignment).​
      disp_height Height of he video stream resolution (used for mapping detection box coordinates and as the original image size for face alignment).
      det_threshold Minimum face detection score,range 0.0~1.0, recommended value: 0.5. Increasing the value reduces false positives but lowers recall; decreasing the value improves recall but increases false positives.​
      box_min_size Minimum face detection box size ,recommended value is 20(pixel)
      face_angle_ratio Head rotation/pitch angle threshold, range 0.0~1.0. Higher values demand stricter frontal angles (recommended value: 0.25).​
      eye_distance Minimum eye distances,recommended value is 20(pixel)
      attr_threshold Score threshold of face attribute recognition, face emotion recognition,range 0.0~1.0,recommended value is 0.5
      motion_sensitive Face movement amplitude threshold,range 0.0~1.0,Higher values demand smaller face movement
      face_quality_threshold Face quality threshold,range 0.0~1.0,recommended value is 0.3

    • Related Data Type and APIs

      ALGO_FR_SetParams

    4.5 FrInput_t

    • Description

      model input attributes structure

    • Definition

      typedef struct
{
    void *p_vir_addr;
    MI_PHY phy_addr;
    MI_U32 buf_size;
    MI_U64 pts;
} FrInput_t;

    • Members

      Member Name Description
      p_vir_addr virtual address of input buffer
      phy_addr physical address of input buffer
      buf_size length of input buffer
      pts timestamp of input buffer

    • Related Data Type and APis

      ALGO_FR_Detect

      ALGO_FR_Attr

      ALGO_FR_FaceQuality

    4.6. FrDetectMode_e

    • Description

      face detection mode

    • Definition

      typedef enum{
    FR_DETECT_MODE_DET_MAX_FACE=0,
    FR_DETECT_MODE_DET_ALL_FACE,
    FR_DETECT_MODE_TRACK_MAX_FACE,
    FR_DETECT_MODE_TRACK_ALL_FACE,
}FrDetectMode_e;

    • Members

      Member Name Description
      FR_DETECT_MODE_DET_MAX_FACE use maximum detected face
      FR_DETECT_MODE_DET_ALL_FACE use all detected faces
      FR_DETECT_MODE_TRACK_MAX_FACE video input,use maximum detected face, add track id
      FR_DETECT_MODE_TRACK_ALL_FACE video input,use all detected faces, add track id

    • Related Data Type and APIs

      ALGO_FR_Detect

    4.7. FrPoint_t

    • Description

      face key-point coordinates structure

    • Definition

      typedef struct FrPoint_
{
    MI_FLOAT x;
    MI_FLOAT y;
}FrPoint_t;

    • Members

      Member Name Description
      x key-point horizontal coordinate
      y key-point vertical coordinate

    • Related Data Type and APIs

      FrBox_t

      ALGO_FR_Detect

      ALGO_FR_Align

      ALGO_FR_FaceQuality

    4.8. FrBox_t

    • Description

      face detection box structure

    • Definition

      typedef struct FrBox_
{
    MI_FLOAT x1;
    MI_FLOAT y1;
    MI_FLOAT x2;
    MI_FLOAT y2;
    MI_FLOAT score;
    FrPoint_t landmark[FR_POINT_LEN];
    MI_U64 face_id;                    // track id
    MI_BOOL befiltered;
    MI_FLOAT move_sensitive;   // lower value means larger face movement
}FrBox_t;

    • Members

      Member Name Description
      X1 x coordinate of face box left-top point
      Y1 y coordinate of face box left-top point
      X2 x coordinate of face box right-bottom point
      Y2 y coordinate of face box right-bottom point
      score face confidence
      landmark 5 face landmarks
      face_id face track id
      befiltered whether face is filtered, true means filtered
      move_sensitive face movement amplitude threshold,range 0.0~1.0,Higher values demand smaller face movement

    • Related Data Type and APIs

      FrPoint_t

      ALGO_FR_Detect

      ALGO_FR_Align

      ALGO_FR_FaceQuality

    4.9. FrAttr_t

    • Description

      face attribute structure

    • Definition

      typedef struct FrAttr_t_
{
    MI_BOOL   eye_glass;
    MI_BOOL   male;
    MI_BOOL   mask;
    MI_S32    age;
    MI_BOOL   mustache;
    MI_S32    emotion;
}FrAttr_t;

    • Members

      Member Name Description
      eye_glass wear glasses or not
      male male or not
      mask wear mask or not
      age age
      mustache have a mustache or not
      emotion emotion type, corresponding relationships are as follows: ​​0: Neutral​​; 1: Happy​​;​2: Sad​​; ​3: Surprised​​;​4: Fearful​​​;​5: Disgusted​​;​6: Angry​

    • Related Data Type and APIs

      ALGO_FR_Attr

    4.10. FaceQualityAngle_t

    • Description

      Face angle estimation output structure

    • Definition

      typedef struct FaceQualityAngle_
{
    MI_FLOAT pitch;
    MI_FLOAT yaw;
    MI_FLOAT roll;
    MI_FLOAT quality;
}FaceQualityAngle_t;

    • Members

      Member Name Description
      pitch face pitch angle
      yaw face yaw angle
      roll face roll angle
      quality face angle Composite Score

    • Related Data Type and APIs

      ALGO_FR_FaceQuality

    5. Error Code

    Error Code Value Description
    E_ALGO_SUCCESS 0 Operation successful
    E_ALGO_HANDLE_NULL 1 Algorithm handle is null
    E_ALGO_INVALID_PARAM 2 Invalid input parameter
    E_ALGO_DEVICE_FAULT 3 Hardware error
    E_ALGO_LOADMODEL_FAIL 4 Model loading failed
    E_ALGO_INIT_FAIL 5 Algorithm initialization failed
    E_ALGO_NOT_INIT 6 Algorithm has not been initialized
    E_ALGO_INPUT_DATA_NULL 7 Algorithm input data is null
    E_ALGO_INVALID_INPUT_SIZE 8 Invalid dimensions of the algorithm input data
    E_ALGO_INVALID_LICENSE 9 Invalid license permission
    E_ALGO_MEMORY_OUT 10 Insufficient memory
    E_ALGO_FILEIO_ERROR 11 File read/write operation error
    E_ALGO_INVALID_OUTPUT_SIZE 12 Invalid dimensions of the algorithm output data
    E_ALGO_INVALID_DECODE_MODE 13 Invalid decode mode
    E_ALGO_MODEL_INVOKE_ERROR 14 Model invoke error
    E_ALGO_INVALID_FILE 15 Invalid file