图像语义分割形象化描述
图像语义分割是指像素级地识别图像，即标注出图像中每个像素所属的对象类别。
目标：一般是将一张RGB图像（height*width*3）或是灰度图（height*width*1）作为输入，输出的是分割图，其中每一个像素包含了其类别的标签（height*width*1）
Unet网络架构
-
Unet的左侧是convolution layers
-
右侧则是upsamping layers
-
convolutions layers中每个pooling layer前输出值会concatenate到对应的upsamping层的输出值中
-
前半部分作用是特征提取，后半部分是上采样。在一些文献中也把这样的结构叫做编码器-解码器结构。
-
上采样部分会融合特征提取部分的输出，这样做实际上是将多尺度特征融合在了一起，以最后一个上采样为例，它的特征既来自第一个卷积block的输出(同尺度特征)，也来自上采样的输出(大尺度特征)
获取原图和分割图路径
all_pics = glob.glob(r'E:HKdatasetHKdataset raining*.png') # 获取所有图片
images = [p for p in all_pics if 'matte' not in p] # 获取原图
annotations = [p for p in all_pics if 'matte' in p] # 获取分割图
制作数据集
np.random.seed(2021)
index = np.random.permutation(len(images))
images = np.array(images)[index]
anno = np.array(annotations)[index]
all_test_pics = glob.glob(r'E:HKdatasetHKdataset esting*.png')
test_images = [p for p in all_test_pics if 'matte' not in p]
test_anno = [p for p in all_test_pics if 'matte' in p]
transform = transforms.Compose([
transforms.Resize((256, 256)),
transforms.ToTensor(),
])
class Portrait_dataset(data.Dataset):
def __init__(self, img_paths, anno_paths):
self.imgs = img_paths
self.annos = anno_paths
def __getitem__(self, index):
img = self.imgs[index]
anno = self.annos[index]
pil_img = Image.open(img)
img_tensor = transform(pil_img)
pil_anno = Image.open(anno)
anno_tensor = transform(pil_anno)
anno_tensor = torch.squeeze(anno_tensor).type(torch.long) # 去掉维数为 1 的维度
anno_tensor[anno_tensor > 0] = 1 # 将分割图转化为只存在0和1两个像素的图像
return img_tensor, anno_tensor
def __len__(self):
return len(self.imgs)
train_dataset = Portrait_dataset(images, anno)
test_dataset = Portrait_dataset(test_images, test_anno)
train_dl = data.DataLoader(train_dataset, batch_size=4, shuffle=True)
test_dl = data.DataLoader(test_dataset, batch_size=4)
定义模型
下采样模型
一个下采样模型包括一层池化+两层卷积
第一层卷积channel的数量由in_channels->out_channels，第二层卷积channel数量由out_channels->out_channels
class Downsample(nn.Module):
def __init__(self, in_channels, out_channels):
super(Downsample, self).__init__()
self.conv_relu = nn.Sequential(
nn.Conv2d(in_channels, out_channels,
kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels,
kernel_size=3, padding=1),
nn.ReLU(inplace=True)
)
self.pool = nn.MaxPool2d(kernel_size=2)
def forward(self, x, is_pool=True):
if is_pool:
x = self.pool(x)
x = self.conv_relu(x)
return x
上采样模型
上采样模型包括两层卷积+一层上采样，上采样采用反卷积
第一层卷积channel的数量由2 * channels->channels，第二层卷积channel数量由channels->channels
上采样再将channel数量减半
class Upsample(nn.Module):
def __init__(self, channels):
super(Upsample, self).__init__()
self.conv_relu = nn.Sequential(
nn.Conv2d(2*channels, channels,
kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(channels, channels,
kernel_size=3, padding=1),
nn.ReLU(inplace=True)
)
self.upconv_relu = nn.Sequential(
nn.ConvTranspose2d(channels,
channels//2,
kernel_size=3,
stride=2,
padding=1,
output_padding=1),
nn.ReLU(inplace=True)
)
def forward(self, x):
x = self.conv_relu(x)
x = self.upconv_relu(x)
return x
模型
模型构成是由5层下采样模型，1个上采样层，3个上采样模型，2层卷积(3*3)，1层卷积(1*1)输出
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.down1 = Downsample(3, 64)
self.down2 = Downsample(64, 128)
self.down3 = Downsample(128, 256)
self.down4 = Downsample(256, 512)
self.down5 = Downsample(512, 1024)
self.up = nn.Sequential(
nn.ConvTranspose2d(1024,
512,
kernel_size=3,
stride=2,
padding=1,
output_padding=1),
nn.ReLU(inplace=True)
)
self.up1 = Upsample(512)
self.up2 = Upsample(256)
self.up3 = Upsample(128)
self.conv_2 = Downsample(128, 64)
self.last = nn.Conv2d(64, 2, kernel_size=1)
def forward(self, x):
x1 = self.down1(x, is_pool=False)
x2 = self.down2(x1)
x3 = self.down3(x2)
x4 = self.down4(x3)
x5 = self.down5(x4)
x5 = self.up(x5)
x5 = torch.cat([x4, x5], dim=1) # 32*32*1024
x5 = self.up1(x5) # 64*64*256)
x5 = torch.cat([x3, x5], dim=1) # 64*64*512
x5 = self.up2(x5) # 128*128*128
x5 = torch.cat([x2, x5], dim=1) # 128*128*256
x5 = self.up3(x5) # 256*256*64
x5 = torch.cat([x1, x5], dim=1) # 256*256*128
x5 = self.conv_2(x5, is_pool=False) # 256*256*64
x5 = self.last(x5) # 256*256*3
return x5