基于飞桨复现语义分割网络HRNet，实现瓷砖缺陷检测

    #获取各阶段的通道数{18,36,72，144}
    channels_2 = cfg.MODEL.HRNET.STAGE2.NUM_CHANNELS
    channels_3 = cfg.MODEL.HRNET.STAGE3.NUM_CHANNELS
    channels_4 = cfg.MODEL.HRNET.STAGE4.NUM_CHANNELS
    #获取各阶段残差单元的循环次数{1,4,3}
    num_modules_2 = cfg.MODEL.HRNET.STAGE2.NUM_MODULES
    num_modules_3 = cfg.MODEL.HRNET.STAGE3.NUM_MODULES
    num_modules_4 = cfg.MODEL.HRNET.STAGE4.NUM_MODULES 
    #步长为2的跨步卷积 f=3*3
    x = conv_bn_layer(input=input,filter_size=3,num_filters=64,stride=2,if_act=True,name='layer1_1')
    #步长为2的跨步卷积 f=3*3
    x = conv_bn_layer(input=x,filter_size=3,num_filters=64,stride=2,if_act=True,name='layer1_2')
    #执行1个残差单元
    la1 = layer1(x, name='layer2')
    #根据输入中最低分辨率特征图生成低分辨率特征图，并规范特征图的通道数
    tr1 = transition_layer([la1], [256], channels_2, name='tr1')
    #执行4次残差卷积，并在每次残差单元结束时进行特征融合
    st2 = stage(tr1, num_modules_2, channels_2, name='st2')
    #根据输入中最低分辨率特征图生成低分辨率特征图，并规范特征图的通道数
    tr2 = transition_layer(st2, channels_2, channels_3, name='tr2')
    #执行3次残差卷积，并在每次残差单元结束时进行特征融合
    st3 = stage(tr2, num_modules_3, channels_3, name='st3')
    #根据输入中最低分辨率特征图生成低分辨率特征图，并规范特征图的通道数
    tr3 = transition_layer(st3, channels_3, channels_4, name='tr3')
    #执行1次残差卷积，并在每次残差单元结束时进行特征融合
    st4 = stage(tr3, num_modules_4, channels_4, name='st4')
    shape = st4[0].shape
    ##获取st4[0]宽高，并进行双线性插值
    height, width = shape[-2], shape[-1]
    st4[1] = fluid.layers.resize_bilinear(st4[1], out_shape=[height, width])
    st4[2] = fluid.layers.resize_bilinear(st4[2], out_shape=[height, width])
    st4[3] = fluid.layers.resize_bilinear(st4[3], out_shape=[height, width])
    #特征通道合并
    out = fluid.layers.concat(st4, axis=1)
    #求总通道数
    last_channels = sum(channels_4)
    #使用1*1卷积进行跨通道的特征融合
    out = conv_bn_layer(input=out,filter_size=1,num_filters=last_channels,stride=1,if_act=True,name='conv-2')
    #使用1*1卷积进行最后的像素分类       
    out = fluid.layers.conv2d(input=out,num_filters=num_classes,filter_size=1,stride=1,padding=0,act=None,
    param_attr=ParamAttr(initializer=MSRA(), name='conv-1_weights'),bias_attr=False)
    #恢复至网络输入的大小
    out = fluid.layers.resize_bilinear(out, input.shape[2:])

pip install -U paddlepaddle-gpu

git clone https://github.com/PaddlePaddle/PaddleSeg

cd PaddleSegpip install -r requirements.txt

python pdseg/check.py --cfg ./configs/hrnet_Magnetic.yaml

PaddleSeg/saved_model/unet_optic/best_model

python pdseg/train.py --use_gpu --cfg ./configs/hrnet_Magnetic.yaml --do_eval

python pdseg/train.py --use_gpu --cfg ./configs/hrnet_Magnetic.yaml --do_eval

[EVAL]#image=81 acc=0.9853 IoU=0.8434
[EVAL]Category IoU: [0.9842 0.7891 0.8468 0.7010 0.9258 0.8136]
[EVAL]Category Acc: [0.9927 0.8871 0.9407 0.9106 0.9597 0.8829]
[EVAL]Kappa:0.9037

python pdseg/vis.py --use_gpu --cfg ./configs/hrnet_Magnetic.yaml

import matplotlib.pyplot as plt
import os
import cv2 
# 定义显示函数
def display(img_name):
    image_dir = os.path.join("./dataset/Magnetic/images", img_name.split(".")[0]+".jpg")
    label_dir = os.path.join("./dataset/Magnetic/color",img_name)
    mask_dir = os.path.join("./visual", img_name)
    img_dir = [image_dir, label_dir, mask_dir]
    plt.figure(figsize=(15, 15))
    title = ['Image', 'label', 'Predict']  
    for i in range(len(title)):
        plt.subplot(1, len(title), i+1)
        plt.title(title[i])
        if i==0:
            img_rgb = cv2.imread(img_dir[i])
        else:
            img = cv2.imread(img_dir[i])
            b,g,r = cv2.split(img)
            img_rgb = cv2.merge([r,g,b])            
        plt.imshow(img_rgb)
        plt.axis('off')
    plt.show()
# 注：第一次运行可能无法显示，再运行一次即可。
img_list=os.listdir("./visual")
for img_name in img_list:    
    display(img_name)