Refactor depth estimation script for clarity and maintainability

[sanowl]

Description

This pull request refactors the Depth Anything V2 script to improve code readability, maintainability, and clarity. The main changes include:

1. Code Structure: The code has been modularized by splitting it into smaller functions for better organization.
2. Comments: Added descriptive comments throughout the code to explain key steps and logic.
3. Logging and Printing: Simplified and enhanced logging for better debugging and monitoring.
4. Function and Variable Names: Improved names for functions and variables to make the code more intuitive.
5. Error Handling: Included error handling for unsupported dataset types.

Changes

• Created functions for:
  • Parsing arguments
  • Setting up logging
  • Initializing training
  • Getting datasets and loaders
  • Initializing the model
  • Setting up the optimizer
  • Training and evaluating the model
• Improved the readability of the main script by breaking it down into logical sections.
• Enhanced logging to provide more detailed and useful information during training and evaluation.

Testing

• The refactored script has been tested on the hypersim and vkitti datasets.
• Verified that the training and evaluation processes run smoothly and produce consistent results.

Additional Notes

• No changes were made to the core algorithm or model architecture, ensuring the same functionality and performance.