【图像去噪】第六期论文复现赛——DnCNN

DnCNN_Paddle    |-- data
         |-- BSD68                  # 测试所用的BSD68数据集
         |-- train                     # 训练所用的数据,包含400张图片
    |-- logs
         |-- net.pdparams        # 训练过程保存的模型参数文件
         |-- vdlrecords.1649579479.log # 完整的训练日志
    |-- test_tipc                     # TIPC: Linux GPU/CPU 基础训练推理测试
    |-- dataset.py                  # 数据及预处理相关代码
    |-- export_model.py       # 预训练模型的导出代码
    |-- LICENSE                     # LICENSE文件
    |-- models.py                  # 模型定义代码
    |-- README.md              # README.md文件
    |-- val.py                         # 模型评估代码
    |-- predict.py                  # 模型预测代码
    |-- train.py                      # 单机单卡训练文件
    |-- train2.py                    # 单机多卡训练文件
    |-- utils.py                       # 一些工具文件
    |-- infer.py                      # 模型推理代码

!pip install scikit-image

class DnCNN(nn.Layer):
    def __init__(self, channels, num_of_layers=17):
        super(DnCNN, self).__init__()
        kernel_size = 3
        padding = 1
        features = 64
        layers = []
        layers.append(nn.Conv2D(in_channels=channels, out_channels=features, kernel_size=kernel_size, padding=padding, bias_attr=False, weight_attr=nn.initializer.KaimingNormal()))

        layers.append(nn.ReLU())        for _ in range(num_of_layers-2):
            layers.append(nn.Conv2D(in_channels=features, out_channels=features, kernel_size=kernel_size, padding=padding, bias_attr=False, weight_attr=nn.initializer.KaimingNormal()))

            layers.append(nn.BatchNorm2D(features, weight_attr=ParamAttr(initializer=nn.initializer.Constant(value=1.)), bias_attr=ParamAttr(initializer=nn.initializer.Constant(value=0.))))
            
            layers.append(nn.ReLU())

        layers.append(nn.Conv2D(in_channels=features, out_channels=channels, kernel_size=kernel_size, padding=padding, bias_attr=False, weight_attr=nn.initializer.KaimingNormal()))

        self.dncnn = paddle.nn.Sequential(*layers)    def forward(self, x):
        y = x
        out = self.dncnn(x)        return y-out

!cd work && python train.py --preprocess True --data_dir data/train --val_dir data/BSD68 --num_of_layers 17 --noiseL 15 --val_noiseL 15

!cd work && python train.py --num_of_layers 17 --noiseL 15 --val_noiseL 15

!cd work && python val.py --log_dir logs --data_path data/BSD68/ --test_noiseL 15

!cd work && python predict.py --log_dir logs --data_path data/BSD68/ --save_path results/ --test_noiseL 15 --save_images

PSNR on test data 31.739193

# 先上传一张图片，import os.path as ospfrom IPython.display import displayfrom PIL import Image
img_path = 'butterfly.png' # 改成自己上传的图片名称full_img_path = osp.join(osp.abspath('work/test_images/'), img_path)
img = Image.open(full_img_path).convert('RGB')print('以下为上传的图片:')
display(img)

!cd work && python predict_single.py --clean_img $full_img_path --save_images --log_dir logs

# 去噪效果查看import globfrom IPython.display import displayfrom PIL import Image

imgs = glob.glob('work/test_images/*')for path in imgs:    print(path)
    img = Image.open(path)
    display(img)