model_utils.py

import torch
import torch.nn as nn
from torchvision import models, transforms
from PIL import Image
import io

# 1. Define the model architecture EXACTLY as in your training script
def create_model(num_classes: int):
    model = models.resnet18(weights=None) # Using weights=None since we are loading our own
    model.fc = nn.Linear(model.fc.in_features, num_classes)
    return model

# 2. Define the EXACT same evaluation transforms from your training script
transform_eval = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406],
                         [0.229, 0.224, 0.225])
])

# 3. Create a function to make a prediction
def predict(model: nn.Module, image_bytes: bytes, class_names: list):
    """
    Takes a model, image bytes, and class names, returns the prediction and confidence.
    """
    # Load image from bytes
    image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
    
    # Preprocess the image
    input_tensor = transform_eval(image).unsqueeze(0) # Add batch dimension
    
    # Make prediction
    model.eval()
    with torch.no_grad():
        output = model(input_tensor)
        probabilities = torch.nn.functional.softmax(output[0], dim=0)
        confidence, predicted_idx = torch.max(probabilities, 0)
    
    predicted_class = class_names[predicted_idx.item()]
    
    return {
        "predicted_id": predicted_class,
        "confidence": confidence.item()
    }