解决YOLOv8模型在不同尺寸图像上预测失败的问题：图像预处理关键技术

import torchvision.transforms as transforms
from PIL import Image
import torch

def preprocess_image_pytorch(image_path: str, desired_size: tuple = (512, 512)) -> torch.Tensor:
    """
    使用PyTorch对图像进行预处理（尺寸调整和转换为张量）。

    Args:
        image_path (str): 图像文件的路径。
        desired_size (tuple): 目标图像尺寸，例如 (宽度, 高度)。

    Returns:
        torch.Tensor: 经过预处理的图像张量，可直接输入YOLOv8模型。
    """
    try:
        image = Image.open(image_path).convert("RGB") # 确保图像是RGB格式
    except FileNotFoundError:
        print(f"错误：找不到图像文件 {image_path}")
        return None
    except Exception as e:
        print(f"加载图像时发生错误：{e}")
        return None

    # 定义图像变换管道
    transform = transforms.Compose([
        transforms.Resize(desired_size), # 调整图像尺寸
        transforms.ToTensor(),           # 将PIL图像转换为PyTorch张量 (HWC -> CHW, 0-255 -> 0.0-1.0)
    ])

    resized_image_tensor = transform(image)
    # YOLOv8模型通常期望批处理维度 (Batch, Channels, Height, Width)
    # 因此，需要添加一个批处理维度
    return resized_image_tensor.unsqueeze(0)

# 示例用法：
# image_path = "path/to/your/large_image.jpg"
# processed_image = preprocess_image_pytorch(image_path, desired_size=(512, 512))

# if processed_image is not None:
#     # 假设您已经加载了YOLOv8模型，例如：
#     # from ultralytics import YOLO
#     # model = YOLO('yolov8n.pt') # 或您训练好的模型
#     # results = model(processed_image)
#     print(f"预处理后的图像张量尺寸：{processed_image.shape}")

import tensorflow as tf
from PIL import Image
import numpy as np

def preprocess_image_tensorflow(image_path: str, desired_size: tuple = (512, 512)) -> tf.Tensor:
    """
    使用TensorFlow对图像进行预处理（尺寸调整和转换为张量）。

    Args:
        image_path (str): 图像文件的路径。
        desired_size (tuple): 目标图像尺寸，例如 (宽度, 高度)。

    Returns:
        tf.Tensor: 经过预处理的图像张量，可直接输入YOLOv8模型。
    """
    try:
        image_pil = Image.open(image_path).convert("RGB") # 确保图像是RGB格式
        # 将PIL图像转换为NumPy数组，再转换为TensorFlow张量
        image_np = np.array(image_pil)
        image_tensor = tf.convert_to_tensor(image_np, dtype=tf.float32)
    except FileNotFoundError:
        print(f"错误：找不到图像文件 {image_path}")
        return None
    except Exception as e:
        print(f"加载图像时发生错误：{e}")
        return None

    # 调整图像尺寸。tf.image.resize期望尺寸为 (高度, 宽度)
    # 并且通常期望输入是浮点数张量 (0-255 或 0.0-1.0)
    # YOLOv8通常期望0-255范围的输入，因此我们不在此处归一化到0-1
    resized_image = tf.image.resize(image_tensor, size=desired_size, method=tf.image.ResizeMethod.BILINEAR)

    # YOLOv8模型通常期望批处理维度 (Batch, Height, Width, Channels)
    # 因此，需要添加一个批处理维度
    return tf.expand_dims(resized_image, axis=0)

# 示例用法：
# image_path = "path/to/your/large_image.jpg"
# processed_image = preprocess_image_tensorflow(image_path, desired_size=(512, 512))

# if processed_image is not None:
#     # 假设您已经加载了YOLOv8模型
#     # from ultralytics import YOLO
#     # model = YOLO('yolov8n.pt') # 或您训练好的模型
#     # results = model(processed_image)
#     print(f"预处理后的图像张量尺寸：{processed_image.shape}")