I'm a beginner when it comes to Oak-D.
I've played around with the demo script quite a bit as well as written some of my own pipelines
I've successfully uploaded a custom trained Mobile-net detection models to the demo script.
However, I've spent the past few days trying to get a faster_rcnn_v2_coco detection model to work in the demo but continue to get the "Couldn't read data from stream: 'color'". I just found this forum from last year:
Was looking to get confirmation that this might be the same issue, and that a resolution is yet to be had on the issue.
In case this isn't the issue in the linked forum, I'm using the custom_model.json text below (uploaded in image because text wasn't formatting properly):
And am using the example handler.py file for mobilenet models with the draw function just 'Pass'ing and with a print statement at the beginning of the decode function confirming the demo never enters the decode function (planned on deciphering the nnet output as soon as it did enter the function).
My setup is
- an Oak-D purchased within the last 6 months
- a fresh download of the depthai demo repository
- running depthai version 18.104.22.168 (required by depthai_demo,py)
- python 3.9.2
- using the USB cable that came in the kit (USB3 I think)
- Oak-d is plugged into 5v power supply that came with kit
Any thoughts you have would be greatly appreciated.