Code refactoring

[DimaBir]

Refactoring Summary

This PR encompasses a major refactoring effort to improve the code structure, readability, and maintainability. Below are the main changes made:

1. Directory Structure:

   • Moved utility functions to dedicated modules for better organization.
   • Introduced a benchmark directory to group all benchmarking-related modules.
   • Separated prediction-related functions into a distinct prediction directory.

2. Benchmarking:

   • Decoupled benchmarking logic from the main script.
   • Each model type (PyTorch, ONNX, OpenVINO, TensorRT) now has its own dedicated benchmarking module.
   • Streamlined the benchmarking process to avoid code duplication.

3. Predictions:

   • Separated prediction functions for better clarity.
   • Addressed issues with data type mismatches during predictions.

4. Utilities:

   • Introduced a utils module to house various helper functions and initializations.
   • Streamlined the model initialization process for different environments (CUDA, ONNX, etc.)

5. Error Handling and Logging:

   • Improved error messages and added more comprehensive logging to help with debugging and performance tracking.

6. Visualization:

   • Enhanced the benchmarking visualization with a clearer and more informative plot.
   • Addressed deprecation warnings related to the Seaborn plotting library.