ImageProcess.py

import torch
import torchvision.transforms as transforms

from torchvision.models import ResNet50_Weights

# 사전 학습된 ResNet 불러오기
from torchvision import models

model = models.resnet50(weights=ResNet50_Weights.DEFAULT)
model.eval()

# 라벨 불러오기 (ImageNet 클래스 라벨)
with open("imagenet_classes.txt") as f:
    labels = [line.strip() for line in f.readlines()]

# 변환 파이프라인 (크기 조정, 텐서 변환, 정규화 등)
transform = 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 classify_image(image):
    # image: PIL Image
    img_t = transform(image)
    batch_t = torch.unsqueeze(img_t, 0)
    out = model(batch_t)
    _, index = torch.max(out, 1)
    return labels[index.item()]