unet实现区域分割

https://github.com/zonghaofan/pig-seg/tree/master/disk_segmentation

网络架构：

# coding:utf-8
import tensorflow as tf
import cv2
import numpy as np
import matplotlib.pyplot as pltimg = cv2.imread('./data/test.png')# cv2.imshow('1.jpg',img)
# cv2.waitKey(0)
img = cv2.resize(img, (1024, 1024))
img = np.array(img).astype(np.float32)
img = img[np.newaxis, ...]
print(img.shape)x_input = tf.placeholder(shape=[None, 1024, 1024, 3], dtype=tf.float32)# x=tf.random_normal(shape=[1,1024,1024,3],dtype=tf.float32)
n_filters = [8, 8]# name=1
def conv2d(x, n_filters, training, name, pool=True, activation=tf.nn.relu):with tf.variable_scope('layer{}'.format(name)):for index, filter in enumerate(n_filters):conv = tf.layers.conv2d(x, filter, (3, 3), strides=1, padding='same', activation=None,name='conv_{}'.format(index + 1))conv = tf.layers.batch_normalization(conv, training=training, name='bn_{}'.format(index + 1))conv = activation(conv, name='relu{}_{}'.format(name, index + 1))if pool is False:return convpool = tf.layers.max_pooling2d(conv, pool_size=(2, 2), strides=2, name='pool_{}'.format(name))return conv, pooldef upsampling_2d(tensor, name, size=(2, 2)):h_, w_, c_ = tensor.get_shape().as_list()[1:]h_multi, w_multi = sizeh = h_multi * h_w = w_multi * w_target = tf.image.resize_nearest_neighbor(tensor, size=(h, w), name='upsample_{}'.format(name))return targetdef upsampling_concat(input_A, input_B, name):upsampling = upsampling_2d(input_A, name=name, size=(2, 2))up_concat = tf.concat([upsampling, input_B], axis=-1, name='up_concat_{}'.format(name))return up_concatdef unet(input):#归一化 -1～1input=(input-127.5)/127.5conv1, pool1 = conv2d(input, [8, 8], training=True, name=1)print(conv1.shape)print(pool1.shape)conv2, pool2 = conv2d(pool1, [16, 16], training=True, name=2)print(conv2.shape)print(pool2.shape)conv3, pool3 = conv2d(pool2, [32, 32], training=True, name=3)print(conv3.shape)print(pool3.shape)conv4, pool4 = conv2d(pool3, [64, 64], training=True, name=4)print(conv4.shape)print(pool4.shape)conv5 = conv2d(pool4, [128, 128], training=True, pool=False, name=5)print(conv5.shape)up6 = upsampling_concat(conv5, conv4, name=6)print('up6', up6.shape)conv6 = conv2d(up6, [64, 64], training=True, pool=False, name=6)print(conv6.shape)up7 = upsampling_concat(conv6, conv3, name=7)print('up7', up7.shape)conv7 = conv2d(up7, [32, 32], training=True, pool=False, name=7)print(conv7.shape)up8 = upsampling_concat(conv7, conv2, name=8)print('up8', up8.shape)conv8 = conv2d(up8, [16, 16], training=True, pool=False, name=8)print(conv8.shape)up9 = upsampling_concat(conv8, conv1, name=9)print('up9', up9.shape)conv9 = conv2d(up9, [8, 8], training=True, pool=False, name=9)print(conv9.shape)final = tf.layers.conv2d(conv9, 1, (1, 1), name='final', activation=tf.nn.sigmoid, padding='same')print('final', final.shape)return finalif __name__ == '__main__':final=unet(x_input)with tf.Session() as sess:sess.run(tf.global_variables_initializer())y_final = sess.run(final, feed_dict={x_input: img})result = y_final[0, ...]print(result.shape)print(result[...,:10])# result=cv2.imread('./2.jpg')# result=cv2.resize(result,(640,640))# print(result)cv2.imshow('1.jpg', result)cv2.waitKey(0)

打印结果：这里打印值有小数，故直接imshow就是输出图，而如果imwrite，查看图片的值全是0，1，轮廓也能看清，只不过不是很清晰。

输入：

输出：截图没有完全

unet实现区域分割相关推荐

Unet网络实现叶子病虫害图像分割
作者|李秋键出品|AI科技大本营(ID:rgznai100) 智能化农业作为人工智能应用的重要领域,对较高的图像处理能力要求较高,其中图像分割作为图像处理方法在其中起着重要作用.图像分割是图像分析的 ...
基于U-Net+残差网络的语义分割缺陷检测
一.介绍基于深度学习的缺陷检测主要集中在场景识别.object detection等方法,近年来,推出了一系列优秀的语义分割模型,比如SegNet.FCN.U-Net等.语义分割模型被广泛的应用到场 ...
Python基于改进Unet的新冠肺炎等级分割系统（源码＆教程）
1.研究背景新冠肺炎给人类带来极大威胁,自动精确分割新冠肺炎CT图像感染区域可以辅助医生进行诊断治疗,但新冠肺炎的弥漫性感染.感染区域形状多变.与其他肺部组织极易混淆等给CT图像分割带来挑战.为此, ...
【计算机视觉】新冠肺炎COVID-19 CT影片阳性检测，感染区域分割，肺部分割，智慧医疗实践，医疗影像处理示例
引言新型冠状病毒肺炎(Corona Virus Disease 2019,COVID-19),简称"新冠肺炎",世界卫生组织命名为"2019冠状病毒病".截止 ...
IterNet——迭代式U-Net变形
IterNet 资源 1 数据集 2 数据增强方式 3 初始设置 4 比较范围 5 核心方法 6 源码资源文章代码:ITerNet 1 数据集 DRIVE:共40张,大小565✖584,具体介绍及 ...
[深度学习论文笔记]Brain tumour segmentation using a triplanar ensemble of U-Nets 基于Unet三平面集成的脑肿瘤分割
Brain tumour segmentation using a triplanar ensemble of U-Nets 基于U网络三平面集成的脑肿瘤分割 Published: May 2021 ...
修改中。。。。【AIChallenger 2018 】【眼底水肿病变区域分割】冠军解决方案
内容完善中... 赛道:眼底水肿病变区域分割队名:965728310 比赛地址:https://challenger.ai/competition/fl2018,答辩视频地址可关注官网最新动态特别 ...
FCN与U-Net语义分割算法
FCN与U-Net语义分割算法图像语义分割(Semantic Segmentation)是图像处理和是机器视觉技术中关于图像理解的重要一环,也是 AI 领域中一个重要的分支.语义分割即是对图像中每一 ...
基于FCN，U-Net的深度学习医学影像分割算法（细胞分割算法）以及传统算法分析
本博文主要分析了基于传统计算机视觉的细胞分割算法和基于深度学习的细胞分割算法.主要针对医学影像分割算法.包括了FCN全卷积网络的细胞分割效果,U-Net细胞分割效果.查阅了采用深度学习分割医学影像的几 ...

unet实现区域分割

unet实现区域分割相关推荐

最新文章

热门文章