Auto labeling does not work with RF_DETR model

🤝 Community Help
1
  • Project Type: Object Detection
  • Operating System & Browser: Windows 11, Chrome
  • Project Universe Link or Workspace/Project ID: hinfelaar-meterkast-detection-dmngz
  • Do you grant Roboflow Support permission to access your Workspace for troubleshooting? (Yes/No): Yes

I am trying to auto lable images with my newest model (93.0% mAP@50). But when I try to use it I get 0 annotations. I have tried to change up the confidence but to no avail. When I use an older version (86.9% mAP@50) it does work. The only difference between the two models is the newer version has one more class and uses RF_DETR object detection (medium) where as the older version uses RF_DETR (medium).

Both models use the same data set. I even trained the latest model V12 (V14) from the checkpoint of V11 (V13).

When i look into the preview of model V12 (V14) it does show predictions higher than 95%

Screenshot of preview model
Screenshot of preview model402×537 64.3 KB

2

Hi @Quint_de_Vries , we’re looking to fix the issue, which is that RF-DETR doesn’t work on the browser (inferencejs), so you don’t get any annotations. We’ll try to fix this asap. For now, could you train & use some different model arch for Auto Labeling?

Thanks, Erik

3

Hi @Quint_de_Vries ,

To follow up, I believe this has since been fixed (using RF-DETR for Auto Labeling). Could you please try again?

Thanks, Erik

4

Hi @erik_roboflow ,

Yes it has been fixed. Thanks for the help!

1 Like