[Feature Request] Add PyTorch 2.0+ Compatibility Guide and Update Dependencies

[ramiredddy]

Description

Thank you for this excellent implementation of your CVPR 2020 paper! The repository has been very useful for the crowd detection community.

Issue

The current codebase was developed with older PyTorch versions (around PyTorch 1.x era based on the last update in 2021). As PyTorch has evolved significantly with 2.0+ releases introducing features like torch.compile() and optimized memory management, users may encounter compatibility issues or miss optimization opportunities.

Suggested Improvements

1. Update dependencies documentation: Add a section specifying tested PyTorch versions and CUDA compatibility
2. Migration guide: Provide guidance for users wanting to use PyTorch 2.0+ features
3. Compatibility notes: Document any known breaking changes or deprecated APIs
4. Optional optimizations: Suggest how to leverage PyTorch 2.0's torch.compile() for potential speed improvements

Benefits

• Helps new users avoid version compatibility issues
• Extends the useful life of this valuable research code
• Enables community to leverage modern PyTorch optimizations
• Reduces duplicate issues about dependency conflicts

Example Structure

A simple addition to README.md:

### Tested Environments
- PyTorch 1.x: Fully tested (original implementation)
- PyTorch 2.0+: Community reported working with minor adjustments

Thank you for considering this enhancement!