关于GCU、沐曦GPGPU、MLU、0卡V100资源4月7日恢复上架的公告>>> 关于共建具身智能开源数据集的倡议>>> 关于云脑任务中统一路径访问方式的公告>>> 关于将启智集群GPU资源迁移至智算集群的公告>>>
jm12138
/
Paddle-CheXNet
mirror of https://github.com.cnpmjs.org/jm12138/Paddle-CheXNet
Not watched
1
0
Fork 0
Code
Releases 0 Wiki Activity
Issues 0 Datasets Cloudbrain
You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
42 Commits
1 Branch
64 MiB
1340 Download
 
 
Branch: main
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'main'
${ noResults }
jm12138 a1859d0e9a update README 2 years ago
chexnet add cam vis 2 years ago
dataset update code 2 years ago
logs Update README.md 2 years ago
pretrained_models Update README.md 2 years ago
vis_cam add cam vis 2 years ago
.gitignore Initial commit 2 years ago
LICENSE Initial commit 2 years ago
README.md update README 2 years ago
eval.py update code 2 years ago
requirements.txt Create requirements.txt 2 years ago
train.py update code 2 years ago
vis_cam.py add cam vis 2 years ago

README.md

Paddle-CheXNet

一、简介

二、复现精度

  • 测试精度：

    The final AUROCs: 
The average AUROC is 0.847
The AUROC of Atelectasis is 0.8292392176490241
The AUROC of Cardiomegaly is 0.9142859299652238
The AUROC of Effusion is 0.8875584024095078
The AUROC of Infiltration is 0.711814559849058
The AUROC of Mass is 0.8647347811493854
The AUROC of Nodule is 0.7921467608091082
The AUROC of Pneumonia is 0.7684077120089262
The AUROC of Pneumothorax is 0.8770951989569952
The AUROC of Consolidation is 0.8160893266094902
The AUROC of Edema is 0.8986931866913855
The AUROC of Emphysema is 0.9302391831183919
The AUROC of Fibrosis is 0.837411708221408
The AUROC of Pleural_Thickening is 0.7962435140282585
The AUROC of Hernia is 0.9311000806021127

  • 精度对比：

    Pathology Wang et al. Yao et al. CheXNet arnoweng/CheXNet Release Model arnoweng/CheXNet Improved Model Paddle-CheXNet
    Atelectasis 0.716 0.772 0.8094 0.8294 0.8311 0.8292
    Cardiomegaly 0.807 0.904 0.9248 0.9165 0.9220 0.9143
    Effusion 0.784 0.859 0.8638 0.8870 0.8891 0.8876
    Infiltration 0.609 0.695 0.7345 0.7143 0.7146 0.7118
    Mass 0.706 0.792 0.8676 0.8597 0.8627 0.8647
    Nodule 0.671 0.717 0.7802 0.7873 0.7883 0.7921
    Pneumonia 0.633 0.713 0.7680 0.7745 0.7820 0.7684
    Pneumothorax 0.806 0.841 0.8887 0.8726 0.8844 0.8771
    Consolidation 0.708 0.788 0.7901 0.8142 0.8148 0.8161
    Edema 0.835 0.882 0.8878 0.8932 0.8992 0.8987
    Emphysema 0.815 0.829 0.9371 0.9254 0.9343 0.9302
    Fibrosis 0.769 0.767 0.8047 0.8304 0.8385 0.8374
    Pleural Thickening 0.708 0.765 0.8062 0.7831 0.7914 0.7962
    Hernia 0.767 0.914 0.9164 0.9104 0.9206 0.9311
    Avg AUROCs 0.841 0.843 0.848 0.847

三、数据集

  • 项目使用的数据集为 dataset
  • ChestX-ray 数据集包含 30,805 名患者的 112,120 张正面视图的X射线图像，以及利用 NLP 从相关放射学报告挖掘的 14 类疾病的图像标签（每个图像可以有多个标签）。
  • 数据集含有 14 类常见的胸部病理，包括肺不张、变实、浸润、气胸、水肿、肺气肿、纤维变性、积液、肺炎、胸膜增厚、心脏肥大、结节、肿块和疝气
  • 下载链接：dataset dataset

四、环境依赖

  • 本项目依赖如下模块：

    scikit-learn
paddlepaddle-gpu
opencv-python
numpy
pillow

  • 可通过如下命令安装依赖：

    $ pip install -r requirements.txt

五、快速使用

  • 同步项目代码

  • 下载数据集并解压至 dataset 文件夹

  • 使用如下命令进行模型训练：

    $ python train.py \
    --data_dir=dataset/images \
    --train_list=dataset/labels/train_list.txt \
    --val_list=dataset/labels/val_list.txt \
    --save_dir=save \
    --batch_size=128 \
    --learning_rate=0.001 \
    --decay_epochs=10,15,18 \
    --decay_factor=0.1 \
    --epoch=20

      Epoch 1/20
  step  10/614 - loss: 0.1650 - AUROC_Atelectasis: 0.5343 - AUROC_Cardiomegaly: 0.5442 - AUROC_Effusion: 0.5442 - AUROC_Infiltration: 0.5400 - AUROC_Mass: 0.4854 - AUROC_Nodule: 0.5043 - AUROC_Pneumonia: 0.5403 - AUROC_Pneumothorax: 0.5507 - AUROC_Consolidation: 0.5129 - AUROC_Edema: 0.5020 - AUROC_Emphysema: 0.5979 - AUROC_Fibrosis: 0.5571 - AUROC_Pleural_Thickening: 0.5663 - AUROC_Hernia: 0.5823 - AUROC_avg: 0.5401 - 3s/step
  step  20/614 - loss: 0.1895 - AUROC_Atelectasis: 0.5719 - AUROC_Cardiomegaly: 0.5691 - AUROC_Effusion: 0.6262 - AUROC_Infiltration: 0.5605 - AUROC_Mass: 0.5333 - AUROC_Nodule: 0.5022 - AUROC_Pneumonia: 0.5410 - AUROC_Pneumothorax: 0.5947 - AUROC_Consolidation: 0.5511 - AUROC_Edema: 0.5663 - AUROC_Emphysema: 0.5629 - AUROC_Fibrosis: 0.4929 - AUROC_Pleural_Thickening: 0.5867 - AUROC_Hernia: 0.6170 - AUROC_avg: 0.5626 - 3s/step
  step  30/614 - loss: 0.1637 - AUROC_Atelectasis: 0.5881 - AUROC_Cardiomegaly: 0.5548 - AUROC_Effusion: 0.6691 - AUROC_Infiltration: 0.5767 - AUROC_Mass: 0.5438 - AUROC_Nodule: 0.5179 - AUROC_Pneumonia: 0.5331 - AUROC_Pneumothorax: 0.6317 - AUROC_Consolidation: 0.5818 - AUROC_Edema: 0.6350 - AUROC_Emphysema: 0.5629 - AUROC_Fibrosis: 0.5629 - AUROC_Pleural_Thickening: 0.6101 - AUROC_Hernia: 0.7026 - AUROC_avg: 0.5908 - 3s/step
  ...

  • 使用如下命令进行模型精度测试（默认使用本项目训练的最佳模型参数）：

    $ python eval.py \
    --data_dir=dataset/images \
    --test_list=dataset/labels/test_list.txt \
    --batch_size=128 \
    --ckpt=pretrained_models/model_paddle.pdparams

      => loading checkpoint
  => loaded checkpoint
  100%|█████████████████████████████████████████| 176/176 [12:56<00:00,  4.41s/it]
  The final AUROCs: 
  The average AUROC is 0.847
  The AUROC of Atelectasis is 0.8292392176490241
  The AUROC of Cardiomegaly is 0.9142859299652238
  The AUROC of Effusion is 0.8875584024095078
  The AUROC of Infiltration is 0.711814559849058
  The AUROC of Mass is 0.8647347811493854
  The AUROC of Nodule is 0.7921467608091082
  The AUROC of Pneumonia is 0.7684077120089262
  The AUROC of Pneumothorax is 0.8770951989569952
  The AUROC of Consolidation is 0.8160893266094902
  The AUROC of Edema is 0.8986931866913855
  The AUROC of Emphysema is 0.9302391831183919
  The AUROC of Fibrosis is 0.837411708221408
  The AUROC of Pleural_Thickening is 0.7962435140282585
  The AUROC of Hernia is 0.9311000806021127

六、代码结构与详细说明

  • 代码结构

    │  train.py # 模型训练脚本
│  eval.py # 模型测试脚本
│  vis_cam.py # CAM 可视化脚本
│  requirements.txt # 依赖环境列表
│
├─chexnet # ChexNet 代码
│      data.py # 数据处理
│      densenet.py # DenseNet
│      model.py # CheXNet Model
│      utility.py # 功能代码
│
├─dataset
│  ├─images # 数据集图像
│  │
│  └─labels # 数据集列表
│          test_list.txt # 测试集
│          train_list.txt # 训练集
│          val_list.txt # 验证集
│
├─logs # 训练 log
│
└─pretrained_models # 预训练模型
        model_paddle.pdparams # 本项目训练的参数文件
        model_torch.pdparams # 转换自参考项目的参数文件

  • 参数说明：

    参数 默认值 说明 适用脚本
    data_dir dataset/images 数据集图片目录 train / eval / vis_cam
    save_dir save 保存目录 train / vis_cam
    train_list dataset/labels/train_list.txt 数据集训练集列表 train
    val_list dataset/labels/val_list.txt 数据集验证集列表 train
    test_list dataset/labels/test_list.txt 数据集测试集列表 eval
    batch_size 128 数据处理批大小 train / eval / vis_cam
    epoch 20 训练轮次 train
    learning_rate 0.001 学习率 train
    decay_epochs 10,15,18 学习率衰减轮次 train
    decay_factor 0.1 学习率衰减因子 train
    ckpt pretrained_models/model_paddle.pdparams 预训练模型路径 eval / vis_cam
    show False 是否展示预览 CAM 图像 vis_cam

七、模型信息

  • 模型的总体信息如下：

    信息 说明
    框架版本 Paddle 2.1.2
    骨干网络 DenseNet 121
    应用场景 多标签分类（胸部 X 光片肺炎检测）
    支持硬件 GPU / CPU

  • 具体的网络结构如下图：

Delete task

Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.

cancel
confirm

No Description

Text Python

Contributors (1)
All

Please read the following content carefully:

Dear OpenI User

Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.

For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》

Agree and continue
Disagree, exit