intersect_dicts() in hubconf.py fix

[glenn-jocher]

Fix for #2609 (comment)

import torch

for ch in 1, 3, 4, 10:
    model = torch.hub.load('ultralytics/yolov5', 'yolov5s', channels=ch)
    results = model(np.array((640, 640, ch)))
    results.print()

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refactoring and cleanup of the intersection logic used in model state dict updates.

📊 Key Changes

• intersect_dicts function moved from utils/torch_utils.py to utils/general.py.
• Updated references to intersect_dicts in hubconf.py and train.py to reflect its new location.
• Now using intersect_dicts during model initialization in hubconf.py to filter checkpoint state_dict before loading.

🎯 Purpose & Impact

• This refactoring improves code modularity and maintainability by placing intersect_dicts in a more general utility file.
• It ensures that only relevant parts of the pretrained model checkpoints are loaded, preventing potential shape mismatches and excluding specific keys when necessary.
• Users will benefit from a more robust and error-resistant model loading process, potentially improving the ease of use and reliability of the YOLOv5 models.