Accept `pathlib.Path` objects for path parameters

[vmiller987]

Search before asking

• I have searched the RF-DETR issues and found no similar feature requests.

Description

Feature Request: Accept pathlib.Path objects for path parameters

Currently, all path-related parameters in RFDETR only accept strings:

from pathlib import Path
from rfdetr import RFDETRNano

# Fails
model = RFDETRNano(pretrain_weights=Path("weights/model.pth"))
# PyDantic ValidationError: Input should be a valid string

# Must convert explicitly:
model = RFDETRNano(pretrain_weights=str(Path("weights/model.pth")))

This is inconvenient when integrating RFDETR with other Python code that uses pathlib.Path throughout.

Accept both str and pathlib.Path for all path-related configuration parameters.

Use case

I did a small test by updating the config class to accept Path:

class RFDETRNanoConfig(RFDETRBaseConfig):
    """
    The configuration for an RF-DETR Nano model.
    """
    out_feature_indexes: List[int] = [3, 6, 9, 12]
    num_windows: int = 2
    dec_layers: int = 2
    patch_size: int = 16
    resolution: int = 384
    positional_encoding_size: int = 24
    pretrain_weights: Optional[str | Path] = "rf-detr-nano.pth"

Prediction with Path objects work correctly after this change:

Dataset classes: ['Test']
Number of dataset classes: 1
Loading images: 100%|██████████| 2/2 [00:00<00:00,  3.15it/s]

Additional

This aligns with modern Python conventions as similar libraries accept pathlib.Path objects.

Are you willing to submit a PR?

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