Incorrect Data Path in YOLOv8 Dataset Configuration

Bug Report: Incorrect Data Path in YOLOv8 Dataset Configuration

Description

While using the Roboflow Python client to download a YOLOv8 dataset for a tennis ball tracking project, I encountered an issue with the generated data.yaml file. The file's paths were incorrectly set, leading to a RuntimeError when attempting to train the YOLO model.

Steps to Reproduce

Download the dataset using the Roboflow Python client:

from roboflow import Roboflow
rf = Roboflow(api_key="YOUR_API_KEY")
project = rf.workspace("denica-tran").project("tennisballtracker")
version = project.version(9)
dataset = version.download("yolov8")

from ultralytics import YOLO
model = YOLO('yolov8n.pt')

dataset_path ="TennisBallTracker-9/data.yaml"
model.train(data=dataset_path, epochs=5)

Observe the downloaded TennisBallTracker-9 folder with the standard YOLOv8 dataset structure.
Check the data.yaml file, which incorrectly references paths for training, validation, and test images.
Expected Behavior
The data.yaml file should correctly point to the respective image directories within the dataset folder.

Actual Behavior

The data.yaml file contains incorrect paths, leading to a RuntimeError when the YOLO model attempts to access the image directories:

RuntimeError: Dataset 'TennisBallTracker-9/data.yaml' error ❌ 
Dataset 'TennisBallTracker-9/data.yaml' images not found ⚠️, missing path '/.../android_figurine/datasets/TennisBallTracker-9/TennisBallTracker-9/valid/images'
Note dataset download directory is '/..../android_figurine/datasets'. You can update this in '/..../Ultralytics/settings.yaml'

TL;DR

The issue might be related to residual settings from a previous project (android_figurine) that seems to interfere with the current dataset configuration. It's unclear why the old dataset path is being referenced.

Temporary Solution

I modified the data.yaml file manually to correct the paths, which allowed the training to proceed successfully:

names:

ankle
court
elbow
eye
hand
head
hip
knee
net
nose
person
player
shoulder
tennis ball
tennis racket
nc: 15
path: /..../TennisBallTracker-9/
test: test/images
train: train/images
val: valid/images

Environment

Python version: 3.12

Reproduziere den Download mit dem Roboflow Python client und dem bereitgestellten YOLOv8-Beispiel und untersuche anschließend die generierte TennisBallTracker-9/data.yaml sowie deren train-, Validierungs- und Testpfade. Vergleiche diese Pfade mit der heruntergeladenen Verzeichnisstruktur und dem gemeldeten Ultralytics-Einstellungspfad; als erledigt gilt die Aufgabe, wenn ein frisch heruntergeladener Datensatz trainiert werden kann, ohne data.yaml manuell zu bearbeiten.