utils/image_processing.py

import torch
from torchvision import transforms
from PIL import Image
import io

# Same transformations used during training
def get_transform():
    return transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
    ])

def process_image(image_bytes):
    """Process an image from bytes to tensor"""
    image = Image.open(io.BytesIO(image_bytes)).convert('RGB')
    transform = get_transform()
    return transform(image).unsqueeze(0)  # Add batch dimension