Loading images on OakD but not using the camera

Tanisha

I am trying to run a segmentation model on OAK-D but I am not using the images from the camera.

I am loading the images from my local and then running my model on OAKD. I am not getting similar results. Is there any preprocessing or post processing I have to do on my images?

jakaskerl

Hi @Tanisha
You have to make sure the W and H are correct and that the frame type matches your image. Set it to interleaved RGB, I think it should work.

Thanks,
Jaka

Tanisha

So my model expects a planar, I set the frame type as frame.setType(dai.RawImgFrame.Type.RGP888p) I still don't get the correct visualization.

Matija

Hey Tanisha ,

Please see here how we read and process the image. You will have to make sure image is in CHW format before sending it to the camera.

Tanisha

So I tried everything but I still can't get the segmentations right. I think the reason might be that I'm using a pytorch deeplabv3plus model while all I've seen in the documentations is the tensorflow one.
I am using deeplabv3plus from this library.
qubvel/segmentation_models.pytorch
import segmentation_models_pytorch as smp

Do you think the architecture could be one reason blob doesn't get the segmentations right?