Migrate detectron2_training.ipynb to use HuggingFace Hub

[mturk24]

Replace S3 URL for labels.pkl with HuggingFace Hub download:

• Add huggingface_hub import to imports cell
• Replace wget S3 command with hf_hub_download() call
• Use Cleanlab/object-detection-tutorial dataset with repo_type="dataset"

This ensures consistency with the main tutorials and eliminates dependency on S3 storage.

[jwmueller]

Suggested change

"This notebook demonstrates how to train a [Detectron2](https://github.com/facebookresearch/detectron2/) model on object detection datasets and produce predictions required to run cleanlab's tutorial on detecting label errors in object detection\u00a0data. Note that this notebook fits the model to an entire training set and produces predictions on a held-out validation set. Thus these predictions are only *out-of-sample* for the validation data, and should ideally *only* be used to find mislabeled images amongst the validation set. To instead find mislabeled images amongst an entire dataset, see the analogous notebook in this folder which uses K-fold cross-validation to produce out-of-sample predictions for every image in the dataset.\n",

"This notebook demonstrates how to train a [Detectron2](https://github.com/facebookresearch/detectron2/) model on object detection datasets and produce predictions required to run cleanlab's tutorial on detecting label errors in object detection data. Note that this notebook fits the model to an entire training set and produces predictions on a held-out validation set. Thus these predictions are only *out-of-sample* for the validation data, and should ideally *only* be used to find mislabeled images amongst the validation set. To instead find mislabeled images amongst an entire dataset, see the analogous notebook in this folder which uses K-fold cross-validation to produce out-of-sample predictions for every image in the dataset.\n",