Benchmark and Model Evaluation Metrics

[sridhs21]

This PR adds performance tracking and evaluation metrics.

Changes

Performance Benchmarking (new module: benchmark.py)

• Tracks hardware specifications (CPU, GPU, RAM, CUDA version)
• Monitors training performance: epoch timing, batch throughput, samples/second
• Records GPU memory usage (current and peak allocation)
• Exports benchmark data to JSON via --benchmark and --benchmark-output flags

Model Evaluation Metrics (new module: eval_metrics.py)

• Evaluates model on both training and validation sets post-training
• Computes standard metrics: Precision, Recall, F1 Score, IoU, Accuracy
• Provides confusion matrix breakdown (TP, FP, FN, TN) per-frame and globally
• Automatic overfitting detection by comparing train vs validation F1 scores
• Exports evaluation results to JSON via --eval-output flag

Integration in XPointMLTest.py

• Import benchmark and evaluation modules
• Pass benchmark tracker through training loop when enabled
• Run evaluation after training completes
• Generate formatted console output and JSON exports for both systems

Benefits

• Monitor hardware utilization and training efficiency
• Quantify model accuracy with standard ML metrics
• Detect overfitting automatically
• Enable reproducible comparisons across runs via JSON exports

[cwsmith]

Great. Thank you.

[cwsmith]

@sridhs21 Thank you. Would you please rebase on main or merge main into this branch?

[cwsmith]

Looks great. Thank you.