OAK-1 Focus Issues

JoshuaAlliet

Hi,

I am working on a ball and goal post recognition project with the OAK-1 camera, running inference using Roboflow.

I would like to know if there are any ways, either through the roboflowoak python module or through the depthai or cv2 modules that would allow me to have a continuous video feed and modify the lens position with the press of a key.

I need this because autofocus with my particular camera isn't too reliable, and I need quite sharp images for optimal inference because when the ball is moving, if the image is already blurry then it becomes really difficult to make detections.

Any help whatsoever is welcome, and thank you in advance.

jakaskerl

Hi JoshuaAlliet
You could add the following logic to https://docs.luxonis.com/projects/api/en/latest/samples/ColorCamera/rgb_camera_control/.

ctrl = dai.CameraControl()
ctrl.setManualFocus(<value between 0 and 255>)
controlQueue.send(ctrl)

source

Hope this helps,
Jaka

JoshuaAlliet

jakaskerl

Thanks for your help.

I just want to confirm that the setManualFocus() method is up to date, because when I run this code on my Rasp Pi 4 connected to my OAK-1, I get that the method does not exist.

Is there any other alternative? If no, is it possible to run a roboflow model using this code?

I have a second question as well.

After downloading my train, valid and test images and their respective annotations from roboflow in ZIP format, how do I then convert that format to GZ in order to use in Google Colab for inference with yolo v4 tiny?

I've done some research but nothing useful has come of it.

Any help is welcome, and thanks in advance.

jakaskerl

Hi JoshuaAlliet
Perhaps you are using an outdated version of depthai? The code works on the latest version.

The tar.gz and .zip are very similar as they both function as compressed archives, they are just more geared towards different systems (unix and windows respectively).
You have to unzip the archive to get the contents inside.

Thanks,
Jaka