Automatic model weight download using HuggingFace Hub

[MihinP]

Currently, PartiNet requires users to manually download model weights and specify their path via command-line arguments. This adds friction to the user experience and can be error-prone.

Proposed Enhancement
Implement automatic model weight downloading using the huggingface_hub library, similar to how PyTorch models can be loaded with automatic downloading.

Suggested Implementation

from huggingface_hub import hf_hub_download
def load_partinet_weights(model_type="denoised_micrographs", cache_dir=None):
    """
    Automatically download and load PartiNet weights from HuggingFace Hub.
    
    Args:
        model_type: Either "denoised_micrographs" or "raw_micrographs"
        cache_dir: Optional custom cache directory
    """
    filename = f"{model_type}.pt"
    weight_path = hf_hub_download(
        repo_id="MihinP/PartiNet",
        filename=filename,
        cache_dir=cache_dir
    )
    return torch.load(weight_path)

Backwards Compatibility

Keep existing manual path specification as an option
If no path provided, automatically download from HuggingFace
Add huggingface_hub as a dependency (or optional dependency)

Additional Considerations

Add progress bar for downloads
Handle offline scenarios gracefully
Document the caching behavior for users