Integration with Roboflow

Roboflow is a tool for building and deploying custom computer vision models. When combined with Luxonis OAK devices, it allows for effective deployment of various models.

Note that we currently only support object detection tasks when deploying Roboflow models on the OAK devices.

Installation

To be able to deploy your models, you will need to install roboflowoak, depthai, and opencv-python:

Command Line

1pip install roboflowoak
2pip install depthai
3pip install opencv-python

Deploying a Model from Roboflow

Deploy your Roboflow object detection model on OAK devices as shown in the example below. Consider that you will need to modify certain parameters to suit your specific model and requirements:

model: Replace this with the ID of your model in Roboflow.
version: Insert the specific version number of your model.
api_key: Use the private API key provided by Roboflow for your account.

Python

1from roboflowoak import RoboflowOak
2import cv2
3import time
4import numpy as np
5
6if __name__ == '__main__':
7    # instantiating an object (rf) with the RoboflowOak module
8    rf = RoboflowOak(model="YOUR-MODEL-ID", confidence=0.05, overlap=0.5,
9    version="YOUR-MODEL-VERSION-#", api_key="YOUR-PRIVATE_API_KEY", rgb=True,
10    depth=True, device=None, blocking=True)
11    # Running our model and displaying the video output with detections
12    while True:
13        t0 = time.time()
14        # The rf.detect() function runs the model inference
15        result, frame, raw_frame, depth = rf.detect()
16        predictions = result["predictions"]
17        # {
18        #     predictions:
19        #     [ {
20        #         x: (middle),
21        #         y:(middle),
22        #         width:
23        #         height:
24        #         depth: ###->
25        #         confidence:
26        #         class:
27        #         mask: {
28        #     ]
29        # }
30        # frame - frame after preprocs, with predictions
31        # raw_frame - original frame from your OAK
32        # depth - depth map for raw_frame, center-rectified to the center camera
33        
34        # timing: for benchmarking purposes
35        t = time.time()-t0
36        print("FPS ", 1/t)
37        print("PREDICTIONS ", [p.json() for p in predictions])
38
39        # setting parameters for depth calculation
40        # comment out the following 2 lines out if you're using an OAK without Depth
41        max_depth = np.amax(depth)
42        cv2.imshow("depth", depth/max_depth)
43        # displaying the video feed as successive frames
44        cv2.imshow("frame", frame)
45    
46        # how to close the OAK inference window / stop inference: CTRL+q or CTRL+c
47        if cv2.waitKey(1) == ord('q'):
48            break

For additional information, refer to the official Roboflow tutorial available here.

ON THIS PAGE

Integration with Roboflow

Installation

Deploying a Model from Roboflow