Uploaded Weights Not Working

🤝 Community Help
1

I’ve fine tuned a Yolo11 l segmentation model using ultralytics version 8.3.40. When I upload the weights, I get an “Oops something went wrong, there might be issues with this model”. The imgsz when training was 1280 - would this cause an issue?

I have also uploaded weights from an rf-detr-segmentation-preview model that was trained on robotflow (copied from one project to another), but it has been processing for 10+ hours so far.

2

I’ve tried many more times and still get the same error everytime.

I have downgraded ultralytics one more version to 8.3.39. I have trained a model with all of the default parameters including img_sz. I also changed the model type in the upload script to yolov11l-seg and yolo11l-seg. It always just gives the oops error.

3

I finally managed to get some models that work so far. I found the roboflow documentation and had to use:
yolo task=detect mode=train model=yolo11l-seg.pt..
and even then, the cli seems to change task=segment automatically.

before I was just running yolo train model=yolo11l-seg.pt… and the model always worked outside of roboflow, but would always give me the oops error in roboflow. I have no idea what is the difference in the model created between those commands though.

4

Hi @a1mm ,

Are you uploading the model through the CLI (i.e. the roboflow command) or by directly hitting one of our POST endpoints?

Thanks,

Grzegorz

5

I’m using the Python sdk

6

Thank you!

Can you share the model ID you used when uploading that resulted in error?

Thanks again,

Grzegorz

7

workspace.deploy_model(
model_type=“yolo11l-seg”,
model_path=“./runs/segment/train6”,
project_ids=[“project-id”],
model_name=“yolo11l-seg”
)

8

The above is what i’m using. It works fine as long as I train using the command specified above, but not the way I have traditionally trained. Both models seem to work fine with ultralytics.

Also, I had tried to upload some rf-detr weights (trained on roboflow in a different project). For model_type, I used rf-detr-seg-preview. It has had the pending status for a week now.

9

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