syntax = "proto3";

package skill.vision;

import "common/common.proto";
import "google/protobuf/struct.proto";

option java_package = "com.cloudminds.harix.skill.vision";
option java_outer_classname = "VisionProto";

enum RecognizeType {
  OCR = 0;
  MONEY = 1;
  CAR_PLATE = 2;
  OBJECT = 3;
  FACE = 4;
  FACE_ATTR = 5;
  CAPTION = 6;
  CLASSIFY = 7;
  FALL = 8;
  COMPARE = 9;
  VENDING = 10;
}

enum PersonStatus {
  COME = 0;
  KEEP = 1;
  LEAVE = 2;
}

message ImageRequest {
  message Body {
    skill.vision.RecognizeType type = 1;
    bool image_from_net = 2;
    bytes image = 3;
    int32 flag = 4;
    skill.vision.PersonStatus status = 5;
    map<string, string> option = 6;
    skill.vision.ImageRequest.RecognizeOption recognize_option = 7;
    repeated bytes images = 8;
  }
  enum RecognizeOption {
    NEVER = 0;
    ALWAYS = 1;
    ASK = 2;
  }
  common.CommonReqInfo common_req_info = 1;
  skill.vision.ImageRequest.Body body = 2;
  common.Extra extra = 3;
}

message ImageResponse {
  message Ocr {
    string text = 1;
  }
  message Money {
    message MoneyInfo {
      string class = 1;
      string class_en = 2;
      float confidence = 3;
      string sample_image = 4;
    }
    map<string, skill.vision.ImageResponse.Money.MoneyInfo> result = 1;
  }
  message CarPlate {
    string plate = 1;
  }
  message Object {
    message Result {
      message Label {
        repeated int32 color = 1;
        string name = 2;
        string name_en = 3;
      }
      message Location {
        repeated int32 bbox = 1;
        repeated int32 color = 2;
        string name = 3;
        string name_en = 4;
      }
      repeated skill.vision.ImageResponse.Object.Result.Label labels = 2;
      repeated skill.vision.ImageResponse.Object.Result.Location location = 3;
    }
    skill.vision.ImageResponse.Object.Result result = 1;
    string tag = 2;
  }
  message Face {
    message FaceObject {
      message Rect {
        int32 x = 1;
        int32 y = 2;
        int32 width = 3;
        int32 height = 4;
      }
      string name = 1;
      skill.vision.ImageResponse.Face.FaceObject.Rect rect = 2;
      float similarity = 3;
    }
    int32 face_num = 1;
    repeated skill.vision.ImageResponse.Face.FaceObject faces = 2;
  }
  message FaceAttr {
    message Attribute {
      float age = 1;
      string emotion = 2;
      string gender = 3;
      skill.vision.ImageResponse.Face.FaceObject.Rect rect = 4;
    }
    repeated skill.vision.ImageResponse.FaceAttr.Attribute result = 1;
  }
  message Caption {
    string caption = 1;
    string caption_en = 2;
  }
  message Classify {
  }
  message Fall {
    message Result {
      float confidence = 1;
      string status = 2;
    }
    skill.vision.ImageResponse.Fall.Result result = 1;
  }
  message Compare {
    message Rect {
      int32 x = 1;
      int32 y = 2;
      int32 width = 3;
      int32 height = 4;
    }
    skill.vision.ImageResponse.Compare.Rect imagefiel1 = 1;
    skill.vision.ImageResponse.Compare.Rect imagefiel2 = 2;
    float similarity = 3;
  }
  message Vending {
    message Result {
      message Position {
        string goodsid = 1;
        string goodsname = 2;
        int32 pointbottom = 3;
        int32 pointleft = 4;
        int32 pointright = 5;
        int32 pointtop = 6;
        float predict = 7;
      }
      map<string, int32> goods = 1;
      string image = 2;
      repeated skill.vision.ImageResponse.Vending.Result.Position position = 3;
    }
    skill.vision.ImageResponse.Vending.Result imagefile1 = 1;
    skill.vision.ImageResponse.Vending.Result imagefile2 = 2;
    skill.vision.ImageResponse.Vending.Result imagefile3 = 3;
    skill.vision.ImageResponse.Vending.Result imagefile4 = 4;
  }
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  string nothing = 10;
  skill.vision.ImageResponse.Ocr ocr = 11;
  skill.vision.ImageResponse.Money money = 12;
  skill.vision.ImageResponse.CarPlate car_plate = 13;
  skill.vision.ImageResponse.Object object = 14;
  skill.vision.ImageResponse.Face face = 15;
  skill.vision.ImageResponse.FaceAttr face_attr = 16;
  skill.vision.ImageResponse.Caption caption = 17;
  skill.vision.ImageResponse.Classify classify = 18;
  skill.vision.ImageResponse.Fall fall = 19;
  skill.vision.ImageResponse.Compare compare = 20;
  skill.vision.ImageResponse.Vending vending = 21;
}

message FaceImage {
  common.CommonReqInfo common_req_info = 1;
  bool image_from_net = 2;
  string faceset_id = 3;
  bytes image = 4;
  int32 flag = 5;
  map<string, string> option = 6;
}

message SingleImage {
  common.CommonReqInfo common_req_info = 1;
  bool image_from_net = 2;
  bytes image = 3;
  int32 flag = 4;
  map<string, string> option = 6;
}

message MultiImage {
  common.CommonReqInfo common_req_info = 1;
  bool image_from_net = 2;
  repeated bytes image = 3;
  int32 flag = 4;
  map<string, string> option = 6;
}

message OcrResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Ocr body = 3;
}

message MoneyResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Money body = 3;
}

message CarPlateResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.CarPlate body = 3;
}

message ObjectResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Object body = 3;
}

message FaceResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Face body = 3;
}

message FaceAttrResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.FaceAttr body = 3;
}

message CaptionResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Caption body = 3;
}

message ClassifyResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Classify body = 3;
}

message FallResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Fall body = 3;
}

message CompareResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Compare body = 3;
}

message VendingResponse {
  common.CommonRspInfo common_rsp_info = 1;
  google.protobuf.Struct detail_message = 2;
  skill.vision.ImageResponse.Vending body = 3;
}

service Vision {
  rpc Recognize (skill.vision.ImageRequest) returns (skill.vision.ImageResponse) {}
  rpc StreamingRecognize (stream skill.vision.ImageRequest) returns (skill.vision.ImageResponse) {}
  rpc DetectOcr (skill.vision.SingleImage) returns (skill.vision.OcrResponse) {}
  rpc DetectMoney (skill.vision.SingleImage) returns (skill.vision.MoneyResponse) {}
  rpc DetectCarPlate (skill.vision.SingleImage) returns (skill.vision.CarPlateResponse) {}
  rpc DetectObject (skill.vision.SingleImage) returns (skill.vision.ObjectResponse) {}
  rpc DetectFace (skill.vision.FaceImage) returns (skill.vision.FaceResponse) {}
  rpc DetectFaceAttr (skill.vision.SingleImage) returns (skill.vision.FaceAttrResponse) {}
  rpc DetectCaption (skill.vision.SingleImage) returns (skill.vision.CaptionResponse) {}
  rpc DetectClassify (skill.vision.SingleImage) returns (skill.vision.ClassifyResponse) {}
  rpc DetectFall (skill.vision.SingleImage) returns (skill.vision.FallResponse) {}
  rpc DetectCompare (skill.vision.MultiImage) returns (skill.vision.CompareResponse) {}
  rpc DetectVending (skill.vision.MultiImage) returns (skill.vision.VendingResponse) {}
}