preprocessing.py

from PIL import Image
import numpy as np
import torch

def preprocess_image_with_pil(image, crop_pct=0.875, do_resize=True, do_normalize=True, size=224, image_mean=[0.485, 0.456, 0.406], image_std=[0.229, 0.224, 0.225]):
    # 如果需要缩放
    if do_resize:
        original_size = min(image.size)
        target_size = int(original_size * crop_pct)
        image = image.resize((target_size, target_size), Image.Resampling.BILINEAR)
    
    # 中心裁剪
    width, height = image.size
    left = (width - size) // 2
    top = (height - size) // 2
    right = left + size
    bottom = top + size
    image = image.crop((left, top, right, bottom))
    
    # 转换为 NumPy 数组并归一化
    image = np.array(image).astype(np.float32) / 255.0
    if do_normalize:
        image = (image - image_mean) / image_std
    
    # 转换为张量格式并增加批次维度
    image = np.transpose(image, (2, 0, 1))
    tensor_image = torch.tensor(image, dtype=torch.float32)
    tensor_image = tensor_image.unsqueeze(0)
    
    return tensor_image