src/data/verify_teacher.py

import torch
import sys
import os
from pathlib import Path

# Add project root to path
sys.path.append(str(Path(__file__).parent.parent.parent))
from src.models.teacher import load_teacher_model

def verify_teacher():
    checkpoint_path = 'checkpoints/resnet50.pt'
    if not os.path.exists(checkpoint_path):
        print(f"Error: {checkpoint_path} not found.")
        return False
    
    print(f"Loading teacher model from {checkpoint_path}...")
    try:
        model = load_teacher_model(checkpoint_path, backbone='resnet50', device='cpu')
        print("Teacher model loaded successfully!")
        
        # Check parameter dtypes
        for name, param in model.named_parameters():
            if param.dtype != torch.float32:
                print(f"Warning: Parameter {name} is {param.dtype}")
            break # Just check the first one
        
        # Test forward pass
        dummy_input = torch.randn(1, 3, 224, 224)
        with torch.no_grad():
            p_logits, y_logits = model(dummy_input)
            print(f"Output shapes: Pitch {p_logits.shape}, Yaw {y_logits.shape}")
            
            p_deg, y_deg = model.get_angles(p_logits, y_logits)
            print(f"Predicted angles (dummy): Pitch {p_deg.item():.2f}, Yaw {y_deg.item():.2f}")
            
        return True
    except Exception as e:
        print(f"Error loading teacher: {e}")
        # Print state dict keys to debug if loading failed due to mismatch
        checkpoint = torch.load(checkpoint_path, map_location='cpu')
        print(f"Keys in checkpoint: {list(checkpoint.keys())[:10]}...")
        return False

if __name__ == '__main__':
    verify_teacher()