Train Model/Create New Version Repeatedly Fails

🤝 Community Help
1

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

  • Project Type: Instance Segmentation

  • Operating System & Browser: Mac

  • Project Universe Link or Workspace/Project ID:

  • **Do you grant Roboflow Support permission to access your Workspace for troubleshooting? (Yes/No): Yes
    **
    I trained a model with about 900 images. Everything worked fine. I then uploaded 255 additional images for additional training. I annotated them, and added them to the same dataset. However, when I try to either Create New Version from the Versions page or Train Model from the dataset page, I keep getting the same error right after the files try to Zip: “Training Failed. An unexpected error occurred while starting training.” No other error log. Additionally, for some reason when setting up training the source images section says 256 images out of 1160 are unannotated:

    On the annotations page, when I click on the dataset folder it shows all the images as annotated.

    These are the model parameters (exact same as the first version):

image
image1032×822 25.2 KB

I do have sufficient credits.

I’ve attempted four times with these parameters and always get the same failure message.

I reported the bug, but no response from Roboflow yet.

I am using MS COCO for the checkpoint rather than the prior version as the first version had some errors I’m trying to overcome.

Thanks!!

2

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