Initial commit

.gitignore

+# ignore map, miou, datasets
+map_out/
+miou_out/
+VOCdevkit/
+datasets/
+Medical_Datasets/
+lfw/
+logs/
+model_data/
+.temp_miou_out/
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/

1.py

+import numpy as np
+import matplotlib.pyplot as plt
+import pywt
+from scipy.fft import fft2, ifft2, fftshift, ifftshift
+import os
+
+
+def fft2c(img):
+    """Centered 2D FFT."""
+    return np.fft.fftshift(np.fft.fft2(np.fft.ifftshift(img)))
+
+
+def ifft2c(kspace):
+    """Centered 2D IFFT."""
+    return np.fft.fftshift(np.fft.ifft2(np.fft.ifftshift(kspace)))
+
+
+def soft_threshold(x, lam):
+    """软阈值操作"""
+    return np.sign(x) * np.maximum(np.abs(x) - lam, 0)
+
+
+def total_variation(x, alpha):
+    """计算图像的总变分"""
+    dx = np.diff(x, axis=0)
+    dy = np.diff(x, axis=1)
+    return alpha * (np.sum(np.abs(dx)) + np.sum(np.abs(dy)))
+
+
+def reconstruct_image(kspace_data):
+    im = fftshift(ifft2(kspace_data, (256, 256)), axes=0)
+    # 取模得到实值图像
+    magnitude_image = np.abs(im)
+    # 归一化图像
+    normalized_image = (magnitude_image - magnitude_image.min()) / (magnitude_image.max() - magnitude_image.min())
+    return normalized_image
+
+
+def gpm_mri_reconstruction(kspace_sampled, mask, lam, num_iters=2000, step_size=0.05, tv_alpha=0.05):
+    """
+    使用梯度投影法（Gradient Projection Method, GPM）进行MRI重建，并引入总变分约束
+    :param kspace_sampled: 采样的k空间数据
+    :param mask: 采样掩码
+    :param lam: 正则化参数
+    :param num_iters: 迭代次数
+    :param step_size: 步长
+    :param tv_alpha: 总变分约束强度
+    :return: 重建的图像
+    """
+    # 初始重建（零填充）
+    img_reconstructed = reconstruct_image(kspace_sampled)
+
+    # 显示初始化图像
+    plt.imshow(np.abs(img_reconstructed), cmap='gray')
+    plt.title('Initial Reconstructed MRI Image')
+    plt.show()
+    plt.imsave('initial_reconstructed_image.png', np.abs(img_reconstructed), cmap='gray')
+
+    for i in range(num_iters):
+        # 计算梯度：数据保真项的梯度
+        kspace_current = fft2c(img_reconstructed)
+        grad_data_fidelity = 2 * ifft2c((kspace_current - kspace_sampled) * mask)
+
+        # 计算梯度：稀疏项的梯度（使用小波变换）
+        coeffs = pywt.wavedec2(img_reconstructed, 'db4', level=3)
+        new_coeffs = []
+        for c in coeffs:
+            if isinstance(c, tuple):
+                c_soft = tuple([soft_threshold(x, lam) for x in c])
+                grad_sparse = [x - c_ for x, c_ in zip(c, c_soft)]
+                new_coeffs.append(grad_sparse)
+            else:
+                c_soft = soft_threshold(c, lam)
+                grad_sparse = c - c_soft
+                new_coeffs.append(grad_sparse)
+        grad_sparse_image = pywt.waverec2(new_coeffs, 'db4')
+
+        # 计算梯度：总变分项的梯度
+        grad_x, grad_y = np.gradient(img_reconstructed)
+        grad_tv = np.sqrt(grad_x ** 2 + grad_y ** 2)
+        grad_tv = tv_alpha * grad_tv
+
+        # 计算总的梯度
+        grad_total = grad_data_fidelity + grad_sparse_image + grad_tv
+
+        # 梯度下降步骤
+        img_temp = img_reconstructed - step_size * grad_total
+
+        # 投影步骤：保证满足k空间采样约束
+        kspace_temp = fft2c(img_temp)
+        kspace_reconstructed = kspace_sampled * mask + kspace_temp * (1 - mask)
+        img_reconstructed = reconstruct_image(kspace_reconstructed)
+
+        if (i + 1) % 10 == 0 or i == 0:
+            print(f"Iteration {i + 1}/{num_iters}")
+
+    return img_reconstructed
+
+
+def main():
+    # 设置当前工作目录
+    current_dir = os.getcwd()
+
+    # 加载完整的k空间数据
+    kspace_path = os.path.join(current_dir, 'fid_20250418_163142.npy')
+    if not os.path.exists(kspace_path):
+        print(f"错误: 找不到文件 {kspace_path}")
+        return
+
+    try:
+        kspace_full = np.load(kspace_path)
+        print(f"成功加载k空间数据: {kspace_path}")
+    except Exception as e:
+        print(f"错误: 加载k空间数据时出错: {e}")
+        return
+
+    # 设置 GPM 参数
+    lam = 0.05  # 正则化参数，可以根据需要调整
+    num_iters = 10000  # 迭代次数
+    step_size = 0.05  # 步长，可以根据需要调整
+    tv_alpha = 0.1  # 总变分约束强度，可以根据需要调整
+
+    # 遍历当前目录下的所有.npy文件，排除'mrd_data.npy'
+    for filename in os.listdir(current_dir):
+        if filename.lower().endswith('.npy') and filename!= 'fid_20250418_163142.npy':
+            mask_path = os.path.join(current_dir, filename)
+            try:
+                mask = np.load(mask_path)
+                print(f"处理文件: {mask_path}")
+            except Exception as e:
+                print(f"错误: 加载文件 {mask_path} 时出错: {e}")
+                continue
+
+            # 检查kspace_full和mask的形状是否匹配
+            if kspace_full.shape!= mask.shape:
+                print(f"警告: 文件 {filename} 的形状与k空间数据不匹配，跳过。")
+                continue
+
+            # 采样k空间
+            kspace_sampled = kspace_full * mask
+
+            # 重建图像
+            try:
+                img_reconstructed = gpm_mri_reconstruction(kspace_sampled, mask, lam, num_iters, step_size, tv_alpha)
+                print(f"成功重建图像: {filename}")
+            except Exception as e:
+                print(f"错误: 重建文件 {filename} 时出错: {e}")
+                continue
+
+            # 构造保存的PNG文件名
+            base_name = os.path.splitext(filename)[0]
+            png_filename = f"{base_name}_gpm_reconstructed_fid_20250417_103329.png"
+            png_path = os.path.join(current_dir, png_filename)
+            plt.imshow(img_reconstructed, cmap='gray')  # 设置为灰度图显示
+            plt.title('Reconstructed MRI Image')
+            plt.show()
+            plt.imsave(png_path, np.abs(img_reconstructed), cmap='gray')
+
+
+if __name__ == "__main__":
+    main()

2.py

+import os
+import cv2
+import numpy as np
+import matplotlib.pyplot as plt
+
+# 输入文件夹路径
+input_folder = r'D:\Users\Desktop\img'
+# 输出文件夹路径
+output_folder = r'D:\Users\Desktop\output_images'
+
+# 如果输出文件夹不存在，则创建它
+if not os.path.exists(output_folder):
+    os.makedirs(output_folder)
+
+# 遍历输入文件夹中的所有文件
+for filename in os.listdir(input_folder):
+    if filename.endswith(('.jpg', '.png', '.jpeg')):
+        # 构建完整的文件路径
+        input_path = os.path.join(input_folder, filename)
+        # 读取图像
+        im2_uint8 = cv2.imread(input_path, cv2.IMREAD_GRAYSCALE)
+
+        # 创建自定义 LUT
+        lut = np.zeros((256, 1, 3), dtype=np.uint8)
+        for i in range(256):
+            if i < 32:
+                # 黑到红渐变
+                lut[i, 0, 2] = 255  # 红色通道
+                lut[i, 0, 0] = 0    # 蓝色通道
+                lut[i, 0, 1] = 0    # 绿色通道
+            else:
+                # 红到黄渐变（红+绿=黄）
+                lut[i, 0, 2] = 255      # 红色通道保持最大
+                lut[i, 0, 1] = (i - 128) * 2    # 绿色通道逐渐增加
+
+        # 转换为 BGR 并应用 LUT
+        pseudo_color1 = cv2.LUT(cv2.cvtColor(im2_uint8, cv2.COLOR_GRAY2BGR), lut)
+
+        # 构建输出文件路径
+        output_filename = os.path.splitext(filename)[0] + '_pseudo_color.jpg'
+        output_path = os.path.join(output_folder, output_filename)
+
+        # 保存处理后的图像
+    #    cv2.imwrite(output_path, pseudo_color1)
+
+        # 使用 matplotlib 显示图像（可选）
+        #plt.imshow(pseudo_color1)
+        #plt.title('Pseudo Color Image')
+        #plt.axis('off')  # 不显示坐标轴
+        #plt.show()
+        # 提取原始文件名（不包含扩展名）
+        base_name = os.path.splitext(filename)[0]
+        # 构建输出文件路径
+        output_filename = f'{base_name}_pseudo_color.png'
+        output_path = os.path.join(output_folder, output_filename)
+
+        # 使用 matplotlib 显示并保存图像
+        plt.imshow(pseudo_color1, cmap='viridis')
+        plt.gca().set_axis_off()
+        plt.savefig(output_path, bbox_inches='tight', pad_inches = 0)
+        plt.close()  # 关闭当前图形，避免图形重叠
+
+
+#0 COLORMAP_AUTUMN
+#1 COLORMAP_BONE
+#2 COLORMAP_JET
+#3 COLORMAP_WINTER
+#4 COLORMAP_RAINBOW
+#5 COLORMAP_OCEAN
+# COLORMAP_SUMMER
+#7 COLORMAP_SPRING
+#8 COLORMAP_COOL
+#9 COLORMAP_HSV
+#10 COLORMAP_PINK
+#11 COLORMAP_HOT
+#pseudo_color = cv2.applyColorMap(im2_uint8, cv2.COLORMAP_HSV)
+#COLORMAP_JET：蓝-青-黄-红渐变（类似热力图）
+#COLORMAP_HOT：黑-红-黄-白渐变
+#COLORMAP_HSV：HSV色彩空间循环
+#其他选项如COLORMAP_BONE, COLORMAP_COOL, COLORMAP_PINK等。
\ No newline at end of file

LICENSE

+MIT License
+
+Copyright (c) 2021 Bubbliiiing
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

-# U-net分割模型
+## Unet：U-Net: Convolutional Networks for Biomedical Image Segmentation目标检测模型在Pytorch当中的实现
+---
 
+### 目录
+1. [仓库更新 Top News](#仓库更新)
+2. [相关仓库 Related code](#相关仓库)
+3. [性能情况 Performance](#性能情况)
+4. [所需环境 Environment](#所需环境)
+5. [文件下载 Download](#文件下载)
+6. [训练步骤 How2train](#训练步骤)
+7. [预测步骤 How2predict](#预测步骤)
+8. [评估步骤 miou](#评估步骤)
+9. [参考资料 Reference](#Reference)
+
+## Top News
+**`2022-03`**:**进行大幅度更新、支持step、cos学习率下降法、支持adam、sgd优化器选择、支持学习率根据batch_size自适应调整。**  
+BiliBili视频中的原仓库地址为：https://github.com/bubbliiiing/unet-pytorch/tree/bilibili
+
+**`2020-08`**:**创建仓库、支持多backbone、支持数据miou评估、标注数据处理、大量注释等。**  
+
+## 相关仓库
+| 模型 | 路径 |
+| :----- | :----- |
+Unet | https://github.com/bubbliiiing/unet-pytorch  
+PSPnet | https://github.com/bubbliiiing/pspnet-pytorch
+deeplabv3+ | https://github.com/bubbliiiing/deeplabv3-plus-pytorch
+
+### 性能情况
+**unet并不适合VOC此类数据集，其更适合特征少，需要浅层特征的医药数据集之类的。**
+| 训练数据集 | 权值文件名称 | 测试数据集 | 输入图片大小 | mIOU | 
+| :-----: | :-----: | :------: | :------: | :------: | 
+| VOC12+SBD | [unet_vgg_voc.pth](https://github.com/bubbliiiing/unet-pytorch/releases/download/v1.0/unet_vgg_voc.pth) | VOC-Val12 | 512x512| 58.78 | 
+| VOC12+SBD | [unet_resnet_voc.pth](https://github.com/bubbliiiing/unet-pytorch/releases/download/v1.0/unet_resnet_voc.pth) | VOC-Val12 | 512x512| 67.53 | 
+
+### 所需环境
+torch==1.2.0    
+torchvision==0.4.0   
+
+### 文件下载
+训练所需的权值可在百度网盘中下载。    
+链接: https://pan.baidu.com/s/1A22fC5cPRb74gqrpq7O9-A    
+提取码: 6n2c   
+
+VOC拓展数据集的百度网盘如下：   
+链接: https://pan.baidu.com/s/1vkk3lMheUm6IjTXznlg7Ng    
+提取码: 44mk   
+
+### 训练步骤
+#### 一、训练voc数据集
+1、将我提供的voc数据集放入VOCdevkit中（无需运行voc_annotation.py）。  
+2、运行train.py进行训练，默认参数已经对应voc数据集所需要的参数了。  
+
+#### 二、训练自己的数据集
+1、本文使用VOC格式进行训练。  
+2、训练前将标签文件放在VOCdevkit文件夹下的VOC2007文件夹下的SegmentationClass中。    
+3、训练前将图片文件放在VOCdevkit文件夹下的VOC2007文件夹下的JPEGImages中。    
+4、在训练前利用voc_annotation.py文件生成对应的txt。    
+5、注意修改train.py的num_classes为分类个数+1。    
+6、运行train.py即可开始训练。  
+
+#### 三、训练医药数据集
+1、下载VGG的预训练权重到model_data下面。  
+2、按照默认参数运行train_medical.py即可开始训练。
+
+### 预测步骤
+#### 一、使用预训练权重
+##### a、VOC预训练权重
+1. 下载完库后解压，如果想要利用voc训练好的权重进行预测，在百度网盘或者release下载权值，放入model_data，运行即可预测。  
+```python
+img/street.jpg
+```    
+2. 在predict.py里面进行设置可以进行fps测试和video视频检测。    
+##### b、医药预训练权重
+1. 下载完库后解压，如果想要利用医药数据集训练好的权重进行预测，在百度网盘或者release下载权值，放入model_data，修改unet.py中的model_path和num_classes；
+```python
+_defaults = {
+    #-------------------------------------------------------------------#
+    #   model_path指向logs文件夹下的权值文件
+    #   训练好后logs文件夹下存在多个权值文件，选择验证集损失较低的即可。
+    #   验证集损失较低不代表miou较高，仅代表该权值在验证集上泛化性能较好。
+    #-------------------------------------------------------------------#
+    "model_path"    : 'model_data/unet_vgg_medical.pth',
+    #--------------------------------#
+    #   所需要区分的类的个数+1
+    #--------------------------------#
+    "num_classes"   : 2,
+    #--------------------------------#
+    #   所使用的的主干网络：vgg、resnet50   
+    #--------------------------------#
+    "backbone"      : "vgg",
+    #--------------------------------#
+    #   输入图片的大小
+    #--------------------------------#
+    "input_shape"   : [512, 512],
+    #--------------------------------#
+    #   blend参数用于控制是否
+    #   让识别结果和原图混合
+    #--------------------------------#
+    "blend"         : True,
+    #--------------------------------#
+    #   是否使用Cuda
+    #   没有GPU可以设置成False
+    #--------------------------------#
+    "cuda"          : True,
+}
+```
+2. 运行即可预测。  
+```python
+img/cell.png
+```
+#### 二、使用自己训练的权重
+1. 按照训练步骤训练。    
+2. 在unet.py文件里面，在如下部分修改model_path、backbone和num_classes使其对应训练好的文件；**model_path对应logs文件夹下面的权值文件**。    
+```python
+_defaults = {
+    #-------------------------------------------------------------------#
+    #   model_path指向logs文件夹下的权值文件
+    #   训练好后logs文件夹下存在多个权值文件，选择验证集损失较低的即可。
+    #   验证集损失较低不代表miou较高，仅代表该权值在验证集上泛化性能较好。
+    #-------------------------------------------------------------------#
+    "model_path"    : 'model_data/unet_vgg_voc.pth',
+    #--------------------------------#
+    #   所需要区分的类的个数+1
+    #--------------------------------#
+    "num_classes"   : 21,
+    #--------------------------------#
+    #   所使用的的主干网络：vgg、resnet50   
+    #--------------------------------#
+    "backbone"      : "vgg",
+    #--------------------------------#
+    #   输入图片的大小
+    #--------------------------------#
+    "input_shape"   : [512, 512],
+    #--------------------------------#
+    #   blend参数用于控制是否
+    #   让识别结果和原图混合
+    #--------------------------------#
+    "blend"         : True,
+    #--------------------------------#
+    #   是否使用Cuda
+    #   没有GPU可以设置成False
+    #--------------------------------#
+    "cuda"          : True,
+}
+```
+3. 运行predict.py，输入    
+```python
+img/street.jpg
+```   
+4. 在predict.py里面进行设置可以进行fps测试和video视频检测。    
+
+### 评估步骤
+1、设置get_miou.py里面的num_classes为预测的类的数量加1。  
+2、设置get_miou.py里面的name_classes为需要去区分的类别。  
+3、运行get_miou.py即可获得miou大小。  
+
+## Reference
+https://github.com/ggyyzm/pytorch_segmentation  
+https://github.com/bonlime/keras-deeplab-v3-plus

cs_Fista.py

+import numpy as np
+import matplotlib.pyplot as plt
+import pywt
+from scipy.fft import fft2, ifft2, fftshift, ifftshift
+import os
+
+
+def fft2c(img):
+    """Centered 2D FFT."""
+    return np.fft.fftshift(np.fft.fft2(np.fft.ifftshift(img)))
+
+
+def ifft2c(kspace):
+    """Centered 2D IFFT."""
+    return np.fft.fftshift(np.fft.ifft2(np.fft.ifftshift(kspace)))
+
+
+def soft_threshold(x, lam):
+    """软阈值操作"""
+    return np.sign(x) * np.maximum(np.abs(x) - lam, 0)
+
+
+def total_variation(x, alpha):
+    """计算图像的总变分"""
+    dx = np.diff(x, axis=0)
+    dy = np.diff(x, axis=1)
+    return alpha * (np.sum(np.abs(dx)) + np.sum(np.abs(dy)))
+
+
+def reconstruct_image(kspace_data):
+    im = fftshift(ifft2(kspace_data, (256, 256)), axes=0)
+    # 取模得到实值图像
+    magnitude_image = np.abs(im)
+    # 归一化图像
+    normalized_image = (magnitude_image - magnitude_image.min()) / (magnitude_image.max() - magnitude_image.min())
+    return normalized_image
+
+
+def fista_mri_reconstruction(kspace_sampled, mask, lam, num_iters=2000, step_size=0.05, tv_alpha=0.05, wavelet='db4'):
+    """
+    使用快速迭代软阈值算法FISTA进行MRI重建，并引入总变分约束
+    :param kspace_sampled: 采样的k空间数据
+    :param mask: 采样掩码
+    :param lam: 正则化参数
+    :param num_iters: 迭代次数
+    :param step_size: 步长
+    :param tv_alpha: 总变分约束强度
+    :param wavelet: 小波基名称，默认为'db4'
+    :return: 重建的图像
+    """
+    # 初始重建（零填充）
+    img_reconstructed = reconstruct_image(kspace_sampled)
+
+    # 显示初始化图像
+    plt.imshow(np.abs(img_reconstructed), cmap='gray')
+    plt.title('Initial Reconstructed MRI Image')
+    plt.show()
+    plt.imsave('initial_reconstructed_image.png', np.abs(img_reconstructed), cmap='gray')
+
+    y = img_reconstructed.copy()
+    t = 1
+
+    for i in range(num_iters):
+        # 将当前图像转换到稀疏域（使用离散小波变换 DWT）
+        coeffs = pywt.wavedec2(y, wavelet, level=3)
+        new_coeffs = []
+        for c in coeffs:
+            if isinstance(c, tuple):
+                c = tuple([soft_threshold(x, lam * step_size) for x in c])
+            else:
+                c = soft_threshold(c, lam * step_size)
+            new_coeffs.append(c)
+
+        # 将稀疏系数转换回图像域
+        img_temp = pywt.waverec2(new_coeffs, wavelet)
+
+        # 引入总变分约束
+        grad_tv = np.gradient(img_temp)
+        grad_tv = np.sqrt(grad_tv[0]**2 + grad_tv[1]**2)
+        img_temp = img_temp - step_size * tv_alpha * grad_tv
+
+        # 数据一致性步骤
+        kspace_reconstructed = fft2c(img_temp)
+        kspace_reconstructed = kspace_sampled * mask + kspace_reconstructed * (1 - mask)
+        img_new = reconstruct_image(kspace_reconstructed)
+
+        t_new = (1 + np.sqrt(1 + 4 * t ** 2)) / 2
+        y = img_new + ((t - 1) / t_new) * (img_new - img_reconstructed)
+
+        img_reconstructed = img_new
+        t = t_new
+
+        if (i + 1) % 10 == 0 or i == 0:
+            print(f"Iteration {i + 1}/{num_iters}")
+
+    return img_reconstructed
+
+
+def main():
+    # 设置当前工作目录
+    current_dir = os.getcwd()
+
+    # 加载完整的k空间数据
+    kspace_path = os.path.join(current_dir, 'fid_20250418_163142.npy')
+    if not os.path.exists(kspace_path):
+        print(f"错误: 找不到文件 {kspace_path}")
+        return
+
+    try:
+        kspace_full = np.load(kspace_path)
+        print(f"成功加载k空间数据: {kspace_path}")
+    except Exception as e:
+        print(f"错误: 加载k空间数据时出错: {e}")
+        return
+
+    # 设置 FISTA 参数
+    lam = 0.05  # 正则化参数，可以根据需要调整
+    num_iters = 10000  # 迭代次数
+    step_size = 0.05  # 步长，可以根据需要调整
+    tv_alpha = 0.1  # 总变分约束强度，可以根据需要调整
+
+    # 定义要尝试的小波基列表
+    #wavelets = ['haar','sym4']
+    wavelets = ['haar']
+    # 遍历当前目录下的所有.npy文件，排除'mrd_data.npy'
+    for filename in os.listdir(current_dir):
+        if filename.lower().endswith('.npy') and filename!= 'fid_20250418_163142.npy':
+            mask_path = os.path.join(current_dir, filename)
+            try:
+                mask = np.load(mask_path)
+                print(f"处理文件: {mask_path}")
+            except Exception as e:
+                print(f"错误: 加载文件 {mask_path} 时出错: {e}")
+                continue
+
+            # 检查kspace_full和mask的形状是否匹配
+            if kspace_full.shape!= mask.shape:
+                print(f"警告: 文件 {filename} 的形状与k空间数据不匹配，跳过。")
+                continue
+
+            # 采样k空间
+            kspace_sampled = kspace_full * mask
+
+            for wavelet in wavelets:
+                # 重建图像
+                try:
+                    img_reconstructed = fista_mri_reconstruction(kspace_sampled, mask, lam, num_iters, step_size, tv_alpha)
+                    print(f"使用{wavelet}小波基成功重建图像: {filename}")
+                except Exception as e:
+                    print(f"错误: 使用{wavelet}小波基重建文件 {filename} 时出错: {e}")
+                    continue
+
+                # 构造保存的PNG文件名，包含小波基名称
+                base_name = os.path.splitext(filename)[0]
+                png_filename = f"{base_name}_{wavelet}_fista_reconstructed_fid_20250417_103329.png"
+                png_path = os.path.join(current_dir, png_filename)
+                plt.imshow(img_reconstructed, cmap='gray')  # 设置为灰度图显示
+                plt.title(f'Reconstructed MRI Image with {wavelet} wavelet')
+                plt.show()
+                plt.imsave(png_path, np.abs(img_reconstructed), cmap='gray')
+
+
+if __name__ == "__main__":
+    main()

cs_admm.py

+import numpy as np
+import matplotlib.pyplot as plt
+from scipy.fftpack import dct, idct
+import os
+
+
+def fft2c(img):
+    """Centered 2D FFT."""
+    return np.fft.fftshift(np.fft.fft2(np.fft.ifftshift(img)))
+
+
+def ifft2c(kspace):
+    """Centered 2D IFFT."""
+    return np.fft.fftshift(np.fft.ifft2(np.fft.ifftshift(kspace)))
+
+
+def generate_sampling_mask(shape, sampling_rate):
+    """生成随机采样掩码"""
+    num_samples = int(sampling_rate * shape[0] * shape[1])
+    mask = np.zeros(shape, dtype=np.float32)
+    indices = np.random.choice(shape[0] * shape[1], num_samples, replace=False)
+    mask.flat[indices] = 1
+    return mask
+
+
+def dct2(a):
+    """二维离散余弦变换"""
+    return dct(dct(a.T, norm='ortho').T, norm='ortho')
+
+
+def idct2(a):
+    """二维离散余弦逆变换"""
+    return idct(idct(a.T, norm='ortho').T, norm='ortho')
+
+
+def soft_threshold(x, lam):
+    """软阈值操作"""
+    return np.sign(x) * np.maximum(np.abs(x) - lam, 0)
+
+
+def admm_mri_reconstruction(kspace_sampled, mask, lam, rho=1.0, num_iters=100):
+    """
+    使用交替方向乘子法进行 MRI 重建
+    :param kspace_sampled: 采样的 k 空间数据
+    :param mask: 采样掩码
+    :param lam: 正则化参数
+    :param rho: 惩罚参数
+    :param num_iters: 迭代次数
+    :return: 重建的图像
+    """
+    # 初始重建（零填充）
+    img_reconstructed = ifft2c(kspace_sampled)
+    z = dct2(img_reconstructed)
+    u = np.zeros_like(z)
+
+    for i in range(num_iters):
+        # 更新 x
+        kspace_x = kspace_sampled * mask + fft2c(ifft2c(idct2(z - u))) * (1 - mask)
+        img_reconstructed = ifft2c(kspace_x)
+
+        # 更新 z
+        sparse_coeff = dct2(img_reconstructed) + u
+        z = soft_threshold(sparse_coeff, lam / rho)
+
+        # 更新 u
+        u = u + (dct2(img_reconstructed) - z)
+
+        if (i + 1) % 10 == 0 or i == 0:
+            print(f"Iteration {i + 1}/{num_iters}")
+
+    return np.abs(img_reconstructed)
+
+
+def main():
+    # 设置当前工作目录
+    current_dir = os.getcwd()
+
+    # 加载完整的 k 空间数据
+    kspace_path = os.path.join(current_dir, 'mrd_data1.npy')
+    if not os.path.exists(kspace_path):
+        print(f"错误: 找不到文件 {kspace_path}")
+        return
+
+    try:
+        kspace_full = np.load(kspace_path)
+        print(f"成功加载 k 空间数据: {kspace_path}")
+    except Exception as e:
+        print(f"错误: 加载 k 空间数据时出错: {e}")
+        return
+
+    # 设置 ADMM 参数
+    lam = 0.1  # 正则化参数，可以根据需要调整
+    rho = 1.0  # 惩罚参数
+    num_iters = 100  # 迭代次数
+
+    # 遍历当前目录下的所有 .npy 文件，排除 'mrd_data.npy'
+    for filename in os.listdir(current_dir):
+        if filename.lower().endswith('.npy') and filename != 'mrd_data.npy':
+            mask_path = os.path.join(current_dir, filename)
+            try:
+                mask = np.load(mask_path)
+                print(f"处理文件: {mask_path}")
+            except Exception as e:
+                print(f"错误: 加载文件 {mask_path} 时出错: {e}")
+                continue
+
+            # 检查 kspace_full 和 mask 的形状是否匹配
+            if kspace_full.shape != mask.shape:
+                print(f"警告: 文件 {filename} 的形状与 k 空间数据不匹配，跳过。")
+                continue
+
+            # 采样 k 空间
+            kspace_sampled = kspace_full * mask
+
+            # 重建图像
+            try:
+                img_reconstructed = admm_mri_reconstruction(kspace_sampled, mask, lam, rho, num_iters)
+                print(f"成功重建图像: {filename}")
+            except Exception as e:
+                print(f"错误: 使用 ADMM 重建文件 {filename} 时出错: {e}")
+                continue
+
+            # 构造保存的 PNG 文件名
+            base_name = os.path.splitext(filename)[0]
+            png_filename = f"{base_name}_ADMM_reconstructed.png"
+            png_path = os.path.join(current_dir, png_filename)
+            plt.imshow(img_reconstructed)
+            plt.axis('off')  # 取消坐标轴
+            plt.tight_layout(pad=0)
+            plt.savefig(png_path, dpi=300)  # 保存为 PNG 文件
+
+
+if __name__ == "__main__":
+    main()
+    
\ No newline at end of file

cs_bregmanpy

+import numpy as np
+import matplotlib.pyplot as plt
+import pywt
+from scipy.fft import fft2, ifft2, fftshift, ifftshift
+import os
+
+
+def fft2c(img):
+    """Centered 2D FFT."""
+    return np.fft.fftshift(np.fft.fft2(np.fft.ifftshift(img)))
+
+
+def ifft2c(kspace):
+    """Centered 2D IFFT."""
+    return np.fft.fftshift(np.fft.ifft2(np.fft.ifftshift(kspace)))
+
+
+def soft_threshold(x, lam):
+    """软阈值操作"""
+    return np.sign(x) * np.maximum(np.abs(x) - lam, 0)
+
+
+def total_variation(x, alpha):
+    """计算图像的总变分"""
+    dx = np.diff(x, axis=0)
+    dy = np.diff(x, axis=1)
+    return alpha * (np.sum(np.abs(dx)) + np.sum(np.abs(dy)))
+
+
+def reconstruct_image(kspace_data):
+    im = fftshift(ifft2(kspace_data, (256, 256)), axes=0)
+    # 取模得到实值图像
+    magnitude_image = np.abs(im)
+    # 归一化图像
+    normalized_image = (magnitude_image - magnitude_image.min()) / (magnitude_image.max() - magnitude_image.min())
+    return normalized_image
+
+
+def split_bregman_mri_reconstruction(kspace_sampled, mask, lam, num_iters=2000, step_size=0.05, tv_alpha=0.05):
+    """
+    使用分裂布雷格曼算法（Split Bregman）进行MRI重建，并引入总变分约束
+    :param kspace_sampled: 采样的k空间数据
+    :param mask: 采样掩码
+    :param lam: 正则化参数
+    :param num_iters: 迭代次数
+    :param step_size: 步长
+    :param tv_alpha: 总变分约束强度
+    :return: 重建的图像
+    """
+    # 初始重建（零填充）
+    img_reconstructed = reconstruct_image(kspace_sampled)
+
+    # 显示初始化图像
+    plt.imshow(np.abs(img_reconstructed), cmap='gray')
+    plt.title('Initial Reconstructed MRI Image')
+    plt.show()
+    plt.imsave('initial_reconstructed_image.png', np.abs(img_reconstructed), cmap='gray')
+
+    # 初始化布雷格曼变量
+    b1 = np.zeros_like(img_reconstructed)
+    b2 = np.zeros_like(img_reconstructed)
+
+    for i in range(num_iters):
+        # 数据一致性步骤
+        kspace_reconstructed = fft2c(img_reconstructed)
+        kspace_reconstructed = kspace_sampled * mask + kspace_reconstructed * (1 - mask)
+        img_temp = reconstruct_image(kspace_reconstructed)
+
+        # 计算总变分
+        grad_tv = np.gradient(img_temp)
+        grad_tv = np.sqrt(grad_tv[0] ** 2 + grad_tv[1] ** 2)
+
+        # 更新图像
+        img_reconstructed = img_temp - step_size * (tv_alpha * grad_tv + b1 + b2)
+
+        # 更新布雷格曼变量
+        b1 = b1 + step_size * (grad_tv - soft_threshold(img_reconstructed + b1, lam))
+        b2 = b2 + step_size * (grad_tv - soft_threshold(img_reconstructed + b2, lam))
+
+        if (i + 1) % 10 == 0 or i == 0:
+            print(f"Iteration {i + 1}/{num_iters}")
+
+    return img_reconstructed
+
+
+def main():
+    # 设置当前工作目录
+    current_dir = os.getcwd()
+
+    # 加载完整的k空间数据
+    kspace_path = os.path.join(current_dir, 'fid_20250418_163142.npy')
+    if not os.path.exists(kspace_path):
+        print(f"错误: 找不到文件 {kspace_path}")
+        return
+
+    try:
+        kspace_full = np.load(kspace_path)
+        print(f"成功加载k空间数据: {kspace_path}")
+    except Exception as e:
+        print(f"错误: 加载k空间数据时出错: {e}")
+        return
+
+    # 设置分裂布雷格曼算法参数
+    lam = 0.05  # 正则化参数，可以根据需要调整
+    num_iters = 10000  # 迭代次数
+    step_size = 0.05  # 步长，可以根据需要调整
+    tv_alpha = 0.1  # 总变分约束强度，可以根据需要调整
+
+    # 遍历当前目录下的所有.npy文件，排除'mrd_data.npy'
+    for filename in os.listdir(current_dir):
+        if filename.lower().endswith('.npy') and filename!= 'fid_20250418_163142.npy':
+            mask_path = os.path.join(current_dir, filename)
+            try:
+                mask = np.load(mask_path)
+                print(f"处理文件: {mask_path}")
+            except Exception as e:
+                print(f"错误: 加载文件 {mask_path} 时出错: {e}")
+                continue
+
+            # 检查kspace_full和mask的形状是否匹配
+            if kspace_full.shape!= mask.shape:
+                print(f"警告: 文件 {filename} 的形状与k空间数据不匹配，跳过。")
+                continue
+
+            # 采样k空间
+            kspace_sampled = kspace_full * mask
+
+            # 重建图像
+            try:
+                img_reconstructed = split_bregman_mri_reconstruction(kspace_sampled, mask, lam, num_iters, step_size, tv_alpha)
+                print(f"成功重建图像: {filename}")
+            except Exception as e:
+                print(f"错误: 重建文件 {filename} 时出错: {e}")
+                continue
+
+            # 构造保存的PNG文件名
+            base_name = os.path.splitext(filename)[0]
+            png_filename = f"{base_name}_split_bregman_reconstructed_fid_20250417_103329.png"
+            png_path = os.path.join(current_dir, png_filename)
+            plt.imshow(img_reconstructed, cmap='gray')  # 设置为灰度图显示
+            plt.title('Reconstructed MRI Image')
+            plt.show()
+            plt.imsave(png_path, np.abs(img_reconstructed), cmap='gray')
+
+
+if __name__ == "__main__":
+    main()

cs_ista.py

+import numpy as np
+import matplotlib.pyplot as plt
+import pywt
+from scipy.fft import fft2, ifft2, fftshift, ifftshift
+import os
+
+
+def fft2c(img):
+    """Centered 2D FFT."""
+    return np.fft.fftshift(np.fft.fft2(np.fft.ifftshift(img)))
+
+
+def ifft2c(kspace):
+    """Centered 2D IFFT."""
+    return np.fft.fftshift(np.fft.ifft2(np.fft.ifftshift(kspace)))
+
+
+def soft_threshold(x, lam):
+    """软阈值操作"""
+    return np.sign(x) * np.maximum(np.abs(x) - lam, 0)
+
+
+def total_variation(x, alpha):
+    """计算图像的总变分"""
+    dx = np.diff(x, axis=0)
+    dy = np.diff(x, axis=1)
+    return alpha * (np.sum(np.abs(dx)) + np.sum(np.abs(dy)))
+
+
+def reconstruct_image(kspace_data):
+    im = fftshift(ifft2(kspace_data, (256, 256)), axes=0)
+    # 取模得到实值图像
+    magnitude_image = np.abs(im)
+    # 归一化图像
+    normalized_image = (magnitude_image - magnitude_image.min()) / (magnitude_image.max() - magnitude_image.min())
+    return normalized_image
+
+
+def ista_mri_reconstruction(kspace_sampled, mask, lam, num_iters=2000, step_size=0.05, tv_alpha=0.05):
+    """
+    使用迭代收缩阈值算法（ISTA）进行MRI重建，并引入总变分约束
+    :param kspace_sampled: 采样的k空间数据
+    :param mask: 采样掩码
+    :param lam: 正则化参数
+    :param num_iters: 迭代次数
+    :param step_size: 步长
+    :param tv_alpha: 总变分约束强度
+    :return: 重建的图像
+    """
+    # 初始重建（零填充）
+    img_reconstructed = reconstruct_image(kspace_sampled)
+
+    # 显示初始化图像
+    plt.imshow(np.abs(img_reconstructed), cmap='gray')
+    plt.title('Initial Reconstructed MRI Image')
+    plt.show()
+    plt.imsave('initial_reconstructed_image.png', np.abs(img_reconstructed), cmap='gray')
+
+    for i in range(num_iters):
+        # 数据一致性步骤
+        kspace_reconstructed = fft2c(img_reconstructed)
+        kspace_reconstructed = kspace_sampled * mask + kspace_reconstructed * (1 - mask)
+        img_temp = reconstruct_image(kspace_reconstructed)
+
+        # 将当前图像转换到稀疏域（使用离散小波变换 DWT）
+        coeffs = pywt.wavedec2(img_temp, 'db4', level=3)
+        new_coeffs = []
+        for c in coeffs:
+            if isinstance(c, tuple):
+                c = tuple([soft_threshold(x, lam * step_size) for x in c])
+            else:
+                c = soft_threshold(c, lam * step_size)
+            new_coeffs.append(c)
+
+        # 将稀疏系数转换回图像域
+        img_reconstructed = pywt.waverec2(new_coeffs, 'db4')
+
+        # 引入总变分约束
+        grad_tv = np.gradient(img_reconstructed)
+        grad_tv = np.sqrt(grad_tv[0] ** 2 + grad_tv[1] ** 2)
+        img_reconstructed = img_reconstructed - step_size * tv_alpha * grad_tv
+
+        if (i + 1) % 10 == 0 or i == 0:
+            print(f"Iteration {i + 1}/{num_iters}")
+
+    return img_reconstructed
+
+
+def main():
+    # 设置当前工作目录
+    current_dir = os.getcwd()
+
+    # 加载完整的k空间数据
+    kspace_path = os.path.join(current_dir, 'fid_20250418_163142.npy')
+    if not os.path.exists(kspace_path):
+        print(f"错误: 找不到文件 {kspace_path}")
+        return
+
+    try:
+        kspace_full = np.load(kspace_path)
+        print(f"成功加载k空间数据: {kspace_path}")
+    except Exception as e:
+        print(f"错误: 加载k空间数据时出错: {e}")
+        return
+
+    # 设置 ISTA 参数
+    lam = 0.05  # 正则化参数，可以根据需要调整
+    num_iters = 10000  # 迭代次数
+    step_size = 0.05  # 步长，可以根据需要调整
+    tv_alpha = 0.1  # 总变分约束强度，可以根据需要调整
+
+    # 遍历当前目录下的所有.npy文件，排除'mrd_data.npy'
+    for filename in os.listdir(current_dir):
+        if filename.lower().endswith('.npy') and filename!= 'fid_20250418_163142.npy':
+            mask_path = os.path.join(current_dir, filename)
+            try:
+                mask = np.load(mask_path)
+                print(f"处理文件: {mask_path}")
+            except Exception as e:
+                print(f"错误: 加载文件 {mask_path} 时出错: {e}")
+                continue
+
+            # 检查kspace_full和mask的形状是否匹配
+            if kspace_full.shape!= mask.shape:
+                print(f"警告: 文件 {filename} 的形状与k空间数据不匹配，跳过。")
+                continue
+
+            # 采样k空间
+            kspace_sampled = kspace_full * mask
+
+            # 重建图像
+            try:
+                img_reconstructed = ista_mri_reconstruction(kspace_sampled, mask, lam, num_iters, step_size, tv_alpha)
+                print(f"成功重建图像: {filename}")
+            except Exception as e:
+                print(f"错误: 重建文件 {filename} 时出错: {e}")
+                continue
+
+            # 构造保存的PNG文件名
+            base_name = os.path.splitext(filename)[0]
+            png_filename = f"{base_name}_ista_reconstructed_fid_20250417_103329.png"
+            png_path = os.path.join(current_dir, png_filename)
+            plt.imshow(img_reconstructed, cmap='gray')  # 设置为灰度图显示
+            plt.title('Reconstructed MRI Image')
+            plt.show()
+            plt.imsave(png_path, np.abs(img_reconstructed), cmap='gray')
+
+
+if __name__ == "__main__":
+    main()

demo.py

+'''
+Author: zhuwanjie 2268677665@qq.com
+Date: 2025-04-21 11:34:18
+LastEditors: zhuwanjie 2268677665@qq.com
+LastEditTime: 2025-04-30 09:56:59
+FilePath: \unet-pytorch\demo.py
+Description: 这是默认设置,请设置`customMade`, 打开koroFileHeader查看配置 进行设置: https://github.com/OBKoro1/koro1FileHeader/wiki/%E9%85%8D%E7%BD%AE
+'''
+import os
+import numpy as np
+import cv2
+from matplotlib import pyplot as plt
+from scipy.fft import fft2, ifft2, fftshift, ifftshift
+
+
+# 从 K 空间数据重建图像
+def reconstruct_image(kspace_data):
+    print("kspace_data shape:", kspace_data.shape) 
+    im = fftshift(ifft2(kspace_data, (256, 256)), axes=0)  # 
+    # 取模得到实值图像
+    magnitude_image = np.abs(im)
+    # 归一化图像
+    normalized_image = (magnitude_image - magnitude_image.min()) / (magnitude_image.max() - magnitude_image.min())
+    # 转换为 8 位无符号整数
+    uint8_image = (normalized_image * 255).astype(np.uint8)
+    # 转换为 RGB 图像
+    rgb_image = cv2.cvtColor(uint8_image, cv2.COLOR_GRAY2RGB)
+    return rgb_image
+
+
+def saveImg(path, img):
+    fig = plt.figure(figsize=(img.shape[1] / 100, img.shape[0] / 100), dpi=130)
+    plt.imshow(img, cmap='gray')
+    plt.axis('off')
+    plt.savefig(path, bbox_inches='tight', pad_inches=0, transparent=True)
+    plt.close(fig)
+
+
+def process_kspace_folder(kspace_folder_path, output_folder_path):
+    if not os.path.exists(output_folder_path):
+        os.makedirs(output_folder_path)
+
+    file_names = [f for f in os.listdir(kspace_folder_path) if f.endswith('.npy')]
+    for file_name in file_names:
+        kspace_file_path = os.path.join(kspace_folder_path, file_name)
+        kspace_data = np.load(kspace_file_path)
+
+        # 重建图像
+        reconstructed_image = reconstruct_image(kspace_data)
+
+        # 保存重建后的图像
+        output_file_name = os.path.splitext(file_name)[0] + '_reconstructed.png'
+        output_file_path = os.path.join(output_folder_path, output_file_name)
+        saveImg(output_file_path, reconstructed_image)
+
+
+if __name__ == "__main__":
+    kspace_folder_path = r'D:\Users\Desktop\fid' # 请替换为实际的k空间数据文件夹路径
+    output_folder_path = r'D:\Users\Desktop\img'  # 请替换为实际的输出文件夹路径
+    process_kspace_folder(kspace_folder_path, output_folder_path)

get_miou.py

+import os
+
+from PIL import Image
+from tqdm import tqdm
+
+from unet import Unet
+from utils.utils_metrics import compute_mIoU, show_results
+
+'''
+进行指标评估需要注意以下几点：
+1、该文件生成的图为灰度图，因为值比较小，按照JPG形式的图看是没有显示效果的，所以看到近似全黑的图是正常的。
+2、该文件计算的是验证集的miou，当前该库将测试集当作验证集使用，不单独划分测试集
+3、仅有按照VOC格式数据训练的模型可以利用这个文件进行miou的计算。
+'''
+if __name__ == "__main__":
+    #---------------------------------------------------------------------------#
+    #   miou_mode用于指定该文件运行时计算的内容
+    #   miou_mode为0代表整个miou计算流程，包括获得预测结果、计算miou。
+    #   miou_mode为1代表仅仅获得预测结果。
+    #   miou_mode为2代表仅仅计算miou。
+    #---------------------------------------------------------------------------#
+    miou_mode       = 0
+    #------------------------------#
+    #   分类个数+1、如2+1
+    #------------------------------#
+    num_classes     = 21
+    #--------------------------------------------#
+    #   区分的种类，和json_to_dataset里面的一样
+    #--------------------------------------------#
+    name_classes    = ["background","aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]
+    # name_classes    = ["_background_","cat","dog"]
+    #-------------------------------------------------------#
+    #   指向VOC数据集所在的文件夹
+    #   默认指向根目录下的VOC数据集
+    #-------------------------------------------------------#
+    VOCdevkit_path  = 'VOCdevkit'
+
+    image_ids       = open(os.path.join(VOCdevkit_path, "VOC2007/ImageSets/Segmentation/val.txt"),'r').read().splitlines() 
+    gt_dir          = os.path.join(VOCdevkit_path, "VOC2007/SegmentationClass/")
+    miou_out_path   = "miou_out"
+    pred_dir        = os.path.join(miou_out_path, 'detection-results')
+
+    if miou_mode == 0 or miou_mode == 1:
+        if not os.path.exists(pred_dir):
+            os.makedirs(pred_dir)
+            
+        print("Load model.")
+        unet = Unet()
+        print("Load model done.")
+
+        print("Get predict result.")
+        for image_id in tqdm(image_ids):
+            image_path  = os.path.join(VOCdevkit_path, "VOC2007/JPEGImages/"+image_id+".jpg")
+            image       = Image.open(image_path)
+            image       = unet.get_miou_png(image)
+            image.save(os.path.join(pred_dir, image_id + ".png"))
+        print("Get predict result done.")
+
+    if miou_mode == 0 or miou_mode == 2:
+        print("Get miou.")
+        hist, IoUs, PA_Recall, Precision = compute_mIoU(gt_dir, pred_dir, image_ids, num_classes, name_classes)  # 执行计算mIoU的函数
+        print("Get miou done.")
+        show_results(miou_out_path, hist, IoUs, PA_Recall, Precision, name_classes)
\ No newline at end of file

image_to_video.py

+import cv2
+import os
+import re
+
+
+def natural_sort_key(s):
+    return [int(text) if text.isdigit() else text.lower() for text in re.split('([0-9]+)', s)]
+
+
+def images_to_video(image_folder, video_name, fps=24, target_width=640, target_height=480):
+    # 支持更多图片格式
+    supported_extensions = ('.png', '.jpg', '.jpeg', '.bmp')
+    images = [img for img in os.listdir(image_folder) if img.endswith(supported_extensions)]
+    images.sort(key=natural_sort_key)
+
+    if not images:
+        print("未找到图片文件。")
+        return
+
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    video = cv2.VideoWriter(video_name, fourcc, fps, (target_width, target_height))
+
+    for image in images:
+        img_path = os.path.join(image_folder, image)
+        try:
+            frame = cv2.imread(img_path)
+            if frame is None:
+                print(f"无法读取图片: {img_path}")
+                continue
+            # 调整图片尺寸
+            resized_frame = cv2.resize(frame, (target_width, target_height))
+            video.write(resized_frame)
+        except Exception as e:
+            print(f"处理图片 {img_path} 时出错: {e}")
+
+    video.release()
+    print(f"视频 {video_name} 已生成。")
+
+
+if __name__ == "__main__":
+    image_folder = r'D:\Users\Desktop\item\unet-pytorch\img'  # 替换为实际的图片文件夹路径
+    video_name = 'output_video.mp4'
+    fps = 1
+    # 设置目标尺寸
+    target_width = 640
+    target_height = 640
+    images_to_video(image_folder, video_name, fps, target_width, target_height)
+    
\ No newline at end of file