🪄

Model Integration Guide

Xtreme1 Point Cloud Object Detection Model Integration Guide

Model Service Interface
Request
POST
 http://ip:port/pointCloud/recognition
Parameters
{
 # A request can contain multiple pieces of data
     "datas": [
     # Each frame of data contains one point cloud file
 {
     "id": 1, # UUID
     "pointCloudUrl": "https://path/to/000001.pcd", # url of the point cloud file
     "imageUrls": 
     [
         "https://path/to/image0/000001.jpg",
         ...
     ],
         "cameraConfigUrl": "https://path/to/camera_config/000001.json"
         }
     ]
}
Responses
{
 "code": "OK",
 "message": "",
 "data": [
 {
    "id": 1, # UUID from request
    "code": "OK",
    "message": "",
    "objects": 
  [
       {
        "label": "CAR", # class
        "confidence": 0.7 # Confidence: (0~1)
            
        "x": 0.0, "y": 0.0, "z": 0.0, # The position of the center point of the box
        "dx": 3, "dy": 2, "dz": 1.5, # size of the box
        "rotX": 0, "rotY": 0, "rotZ": 0.5, # The rotation direction of the box (Euler angle XYZ, intrinsic rotations)
        }
  ],
     "confidence": 0.8 # Confidence of the whole data (optional)
       }
 ]
}
Development Services
Point Cloud Object Detection Service
Write a service script
An example: pcdet_demo_service.py
from flask import Flask, request, jsonify
​
app = Flask(__name__)
​
def predict(pcd_url):
     # 1. Download the pcd file via pcd_url (make sure the algorithm service is connected to the Xtreme1 service machine)
     # r = requests. get(pcd_url, allow_redirects=True)
     # 2. Load the pcd file and get the point cloud information
     # pcd_file = io.BytesIO(r.content)
     #pc = load_pcd(pcd_file)
     # 3. Predict and get the result of target detection
     # results = object_detection(pc)
     return [
         # return a detection example
     {
         "label": "CAR", # category
         "confidence": 0.7, # Confidence: (0~1)
        
         "x": 0.0, "y": 0.0, "z": 0.0, # the position of the center point of the box
         "dx": 3, "dy": 2, "dz": 1.5, # size of the box
         "rotX": 0, "rotY": 0, "rotZ": 0.5, # The rotation direction of the box (Euler angle XYZ, intrinsic rotations)
     }
     ]
​
@app.route("/pointCloud/recognition", methods = ['POST'])
def recognition():
     datas = request.json['datas']
​
     result = {
         "code": "OK",
         "message": "",
         "data": [
             {
                 "id": data['id'],
                 "code": 0,
                 "message": "",
                 "objects": predict(data['pointCloudUrl'])
             }
             for data in data
         ]
     }
     return jsonify(result)y
Integration Services
Start Service
# flask 2.2.3
flask --app=pcdet_demo_service run --host=0.0.0.0 --port=5000
Testing Service
curl -X POST http://<ip>:<port>/pointCloud/recognition -H "Content-Type: application/json"  -d '{"datas": [{"id":1, "pointCloudUrl":"http://test"}]}'
Note: replace <ip> and <port>
Integration Services
Log into Xtreme1 platform;
Enter Models/Create My Model, select Data Type as Lidar, and click the Save button to create the model;
Click the newly created model to enter the details page, and click Settings in the panel;
Click the Edit button in the upper right corner
Fill in the service address in the URL edit box:
http://<ip>:<port>pointCloud/recognition
Note: You need to ensure that the machine where the Xtreme1 service is deployed can access this address.
Click the Test Connection button, if successful, it will return the correct response.
{
  "code": "OK",
  "message": "",
  "data": [
    {
      "code": 0,
      "id": 3091721,
      "message": "",
      "objects": [{...}]
    }
  ]
  }
​
​
​

DEVELOPER REFERENCE - Previous

API Document

Next - Changelog

V 0.7

Last modified 3mo ago