data.yaml file has different references for image paths

When a dataset is downloaded via the API, the data.yaml file has inconsistent image paths. For example they look like:

test: ../test/images
train: project-name/train/images
val: project-name/valid/images

Oddly enough, when exported and downloaded manually from the roboflow page, the paths are correct. It would seem that this problem must be two fold:

• data.yaml file is inconsistent
• the Dataset object or the yolo tool must have an issue to compensate for the incorrect path

This was most obvious to me when I changed the name of the downloaded dataset folder and the YOLO training would no longer work.

Simply doing this:

from roboflow import Roboflow
rf = Roboflow(api_key="YOUR_API_KEY")
project = rf.workspace("roboflow-jvuqo").project("football-players-detection-3zvbc")
dataset = project.version(1).download("yolov8")

will illustrate the problem.

Run the provided Roboflow API download example and inspect the generated data.yaml paths. Trace the Dataset download flow and the YOLO training entry point to determine whether the inconsistency is produced by the downloaded file or compensated for by training. Done means API downloads use consistent, relocatable image paths and training still works after the dataset folder is renamed.