Question about dataset discrepancy in Roboflow: Unannotated images and training split size

Describe your question/issue here! (delete this when you post)

  • Project Type:
  • Operating System & Browser:
  • Project Universe Link or Workspace/Project ID:

I’m currently using the starter plan and working with Roboflow. After performing auto-labeling using autodistill locally, I uploaded the images to Roboflow workspace for label adjustments and moved them to the dataset using ‘add dataset’.

When trying to train the model, I noticed something unusual: While I processed 3,540 images in total, some are marked as ‘unannotated’, and the split only shows around 1,200 images.

Could you explain:

  1. What does ‘unannotated’ mean in this context?
  2. Why does the split only contain approximately 1,200 images instead of the full 3,540?

Thank you!

Does the background image fall out naturally?

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.