Semantic Segmentation 基于改进的ResNet-U-Net

大学牲的CV课程实验🥲，基于ResNet改进U-Net，深度监督、融合注意力机制和ASPP模块，实现多类别语义分割。

As a student,it's my CV course experiment 🥲, based on ResNet to improve U-Net, in-depth supervision, integration of attention mechanism and ASPP module, to achieve multi-category semantic segmentation.

特点 feature

• 基于ResNet的编码器，增强特征提取能力

ResNet-based encoder enhances feature extraction ability

• 引入注意力机制，聚焦重要特征区域

Introduce attention mechanisms and focus on important feature areas

• 集成ASPP模块，提升多尺度目标处理能力

Integrate ASPP modules to improve multi-scale target processing ability

• 深度监督，学习多层次特征

In-depth supervision, learning multi-level characteristics

• 图像增强 Image enhancement

文件结构 Files

├── ResNet_Semantic_Segmentation.ipynb   # 主代码文件，包含模型构建与训练流程
├── voc-pascal-2012-segmentation         # 数据集文件夹
│     ├── JPEGImages                     # 数据集图片文件夹
      ├── mask                           # mask图片文件夹
      ├── train.txt                      # 训练集索引
      ├── valid.txt                      # 验证集索引
├── README.md                            # 项目说明文档
└── best.pth                             # 模型                           

大致网络结构 Rough network structure

成果demo

训练50次

训练100次

环境依赖 Environment

• Python 3.x
• Jupyter Notebook
• PyTorch >= 1.7
• torchvision
• numpy
• matplotlib
• ...