using oak-d as camera and running yolov7 with spatial on pc

AbdelrahmanHany

I am new to depthai and i need a little help,i have an oak-d camera and i want to use the oak-d as camera only and run yolov7 with spatial on my pc can anyone help me? i cant seem to find a tutorial.

erik

Hi AbdelrahmanHany ,
I would suggest following this notebook about the deployment of the model to the OAK camera (at the bottom).
Thanks, Erik

YuxiangGuan

erik

Hi Erik,

When I ran the Googld Colab (YoLo V7),

During !python train.py , I am keeping getting this error: AttributeError: module 'numpy' has no attribute 'int'

Any insights or suggestions would be much appreciated!

AbdelrahmanHany

erik hey erik, thanks for replying, i think i didnt make myself clear enough i am not trying to use the on board vpu to run the nn , i want to use my pc using pytorch framework to run yolov7 and i also want to run depth perception/spatial also on my pc , i just want to use the camera as a camera/ to input the video to my code on pc any ideas?

YuxiangGuan

https://colab.research.google.com/github/luxonis/depthai-ml-training/blob/master/colab-notebooks/YoloV7_training.ipynb#scrollTo=cUtWuSlBS-BU Here's the link to the Google Colab

erik

YuxiangGuan Could you please create an issue on depthai-ml-training repo about this numpy issue?
@AbdelrahmanHany oh sorry, makes sense. You could combine video stremaing and depth streaming demos to achieve that. Thoughts?
Thanks, Erik

YuxiangGuan

erik Will do, thanks Erik. Merry Christmas and happy holidays!

AbdelrahmanHany

erik hey Erik , merry Christmas, your solution is working but i am facing a problem , the output of the streaming rgb camera is frames which works fine for yolov7-tiny.pt but when i use yolov7.pt it doesn't work, using the torch.load(yolov7) line of code, (the output video from the yolov7.pt doesn't have any detections), so 😂i am not really sure what's wrong. Thanks

erik

Hi AbdelrahmanHany , I am not familiar with pytorch and it would be best if you asked such question on pytorch-related sites.
Thanks, Erik