UnboundLocalError: local variable 'image_id' referenced before assignment

I annotated and created an augmented dataset using Roboflow. The augmented images and annotations in COCO csv format were exported as a zip file. While training an instance segmentation model using MaskRCNN and OpenCV, I get the below error:

“UnboundLocalError: local variable ‘image_id’ referenced before assignment”

I labeled the data set earlier with another third-party annotator. I did not get any errors while training back then.

  • Project Type: Instance Segmentation
  • Operating System & Browser: Linux/ VS Code

Hi @Md_Navid_Akbar

Was this done using the Detectron2 training notebook or something else?

Hi @leo,

I have been using a variant of this notebook.
I am using TensorFlow 2.0.
This is a sample json for the validation set exported from Roboflow.

Hi @Md_Navid_Akbar

That looks like a notebook that isn’t made by Roboflow. We generally don’t provide support on the forum regarding notebooks that aren’t made by us.

You could try filing an issue on their GitHub or using one of our many notebooks, like the Detectron2 RCNN training notebook I mentioned before.