[Bug] YOLODataset Initialization Crash on Missing 'test' Split

[karttikjangid]

While using the standard COCO8 dataset to verify an object detection pipeline, I found that the YOLODataset loader crashes during initialization if the data.yaml does not explicitly define a path for every split (specifically the test split).

In the YOLO ecosystem, many datasets (including COCO8 and various custom subsets) only provide train and val sets. Currently, the loader attempts to perform an os.path.join on the split paths without verifying if the keys exist in the YAML, leading to an unhandled TypeError.

Steps to Reproduce

The crash can be reproduced in any environment using the following minimal script with the standard COCO8 dataset:

import os
from perceptionmetrics.datasets.yolo import YOLODataset

# 1. Using the standard COCO8 dataset (missing 'test' split in YAML)
# Download: !wget -q [https://github.com/ultralytics/hub/raw/main/example_datasets/coco8.zip](https://github.com/ultralytics/hub/raw/main/example_datasets/coco8.zip)

# 2. Attempt to initialize the dataset
# This triggers the TypeError inside the constructor
dataset = YOLODataset("coco8/coco8.yaml", "coco8")

[karttikjangid]

I have verified this behavior in a clean Google Colab environment using the latest source code from the master branch.

Verified Dataset: COCO8 (coco8.yaml)

Environment: Python 3.12

Root Cause: Lack of a null-check for optional YAML keys (train, val, test) before executing os.path.join.

[karttikjangid]

[karttikjangid]

@dpascualhe Should i start fixing the bug and raise a pr ?

[dpascualhe]

Sure thing! Also, it would be nice if the loader warned the user about missing splits

[dpascualhe]

And good catch! Validating the pipeline with new data is a great step towards robustness. Thanks for that 👍

[karttikjangid]

Hi @dpascualhe,

I have just opened PR #399 to address this. I implemented the safe path resolution using .get() and added the logging.warning for missing/null splits exactly as you suggested!

I also included a new test suite in tests/test_yolo.py to prevent any future regressions on this. Looking forward to your review, and please let me know if you'd like any adjustments!