Roboflow - ultralytics error with dataset path

[robmarkcole]

Search before asking

• I have searched the Roboflow Notebooks issues and found no similar bug report.

Notebook name

Yolov8 custom dataset, but using the python api. Dataset path is apparently incorrect. I know I've had this issue in the past

Bug

I download the dataset

rf = Roboflow(api_key="xx")
project = rf.workspace("org").project("simulated-data")
version = project.version(1)
dataset = version.download("yolov8")

# validate location
dataset.location == '/teamspace/studios/this_studio/simulated-data'

# Now I should be able to train with 
 # Loading a pretrained model
model = YOLO('yolov8l.pt')

# Training the model
model.train(
    data = f'{dataset.location}/data.yaml', # this is correct
...
)

But I get an error:

RuntimeError: Dataset '/teamspace/studios/this_studio/simulated-data/data.yaml' error ❌ 

Dataset '/teamspace/studios/this_studio/simulated-data/data.yaml' images not found ⚠️, 

missing path '/teamspace/studios/this_studio/simulated-data/simulated-data/valid/images'

Why is it looking in a nested directory?

Environment

Lightning studio

Minimal Reproducible Example

See above

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[venkatram-dev]

By default the default datasets directory is set to 'datasets_dir': str(root / 'datasets'),

https://github.com/roboflow/ultralytics-roboflow/blob/main/ultralytics/yolo/utils/__init__.py#L365

Please create this folder datasets like below , navigate to that folder and then download the data .
Below is a sample from google colab notebook.

import os
HOME = os.getcwd()
print(HOME)

!mkdir {HOME}/datasets
%cd {HOME}/datasets

After that download

rf = Roboflow(api_key="xx")
project = rf.workspace("org").project("simulated-data")
version = project.version(1)
dataset = version.download("yolov8")

optional step : validate that data is downloaded under datasets directory

!ls -ltr $HOME/datasets

dataset.location will point to a directory like simulated-data-1 under $HOME/datasets

Run Training

 # Loading a pretrained model
model = YOLO('yolov8l.pt')

# Training the model
model.train(
    data = f'{dataset.location}/data.yaml', # this is correct
...
)

[robmarkcole]

@venkatram-dev I've followed your suggested approach and encountered the same error

[venkatram-dev]

Oh interesting.
I used roboflow-1.1.44 and ultralytics==8.0.196
Please try to use the same versions and see if you still get the errors.

Also, can you share how was the data uploaded and annotated?

Since I am not able to reproduce ,to debug further, Please share a google colab with exact code and error.

[robmarkcole]

I downgraded to those specific versions and now no error!
These are object detection annotations as yolov8 format from Roboflow

[LinasKo]

We've now verified which versions are required to train, validate, deploy and infer with yolov8. Notebooks are updated - let us know if there's further issues!