I’m creating a new dataset of 300ish images of a specific weed and flower. When trying to use the auto-label I have three defined classes that show up very well in the test images, but when I run Auto-Label, all of my images are eventually left “unannotated”. I tried adding less classes, more classes, different descriptions for the classes, and less images to annotate, but the auto-label model always finds “zero objects” when running through my images. Any suggestions on how to fix this? I added a screenshot below to show how well the model is picking up the objects I want.
Hello,
I have the exact same issue, did you figure our of to fix it ?
Best regards
No, I even tried to make my own model by manually labeling images, then use that model to label more images. Same issue, zero annotations.
Hi there, I haven’t been able to reproduce this issue with our Auto Label tool. Can you describe a bit more about the error? Does the job complete “successfully” with images marked as null? – this could indicate the confidence levels are too high?
Have you experienced this issue with Grounding Dino, or with a custom trained model? Would you mind sharing a screenshot of what you mean by “unannotated”?
Thank you for your help debugging this issue!
The job completes “successfully” with the images left unannotated. It doesn’t assign them to null, just says the job is complete. This is with the Ground DINO model, though I get the same results when I use my own custom model.
Also to add more info, the confidence levels are typically very low, 10-30%, when Im trying to use this tool. I thought it could be a background confusion issue, so I tried adding in the “grass” class to see if that could fix it, but it provides the same results with or without this class added.
![Screenshot 2024-10-23 144509|690x451](upload://6cf08g7EldLehdM1mQqPOAiFAFH.jpeg
Thank you for the information! We’re investigating this issue and will report back any updates we find!
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.