首页 开源 资讯 活动 开源许可证
软件工程云服务
软件代码质量检测云服务 持续集成与部署云服务 社区个性化内容推荐服务 贡献审阅人推荐服务 群体化学习服务 重睛鸟代码扫描工具
Watch 1 Star 0 Fork 1

王其/MRI_Deep_learning

forked from chenmingling/MRI_Deep_learning 
代码 Issues 0 Pull Requests 0 Wiki 0 统计
加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目，私有仓库也完全免费 ：）
免费加入
文件
该仓库未声明开源许可证文件（LICENSE），使用请关注具体项目描述及其代码上游依赖。
项目仓库所选许可证以仓库主分支所使用许可证为准
克隆/下载
for_crop_test.py 12.29 KB
一键复制 编辑 原始数据 按行查看 历史
chris1992212 提交于 2018-10-11 15:00 . SR_crop
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197
'''

2-D Convolutional Neural Networks using TensorFlow library for MR Image reconstruction.



'''

import numpy as np

from scipy.misc import imsave, imread, imresize

import tensorflow as tf

import SRCNN_models

from data_inputs import *

from SRCNN_configs import *

os.environ['CUDA_VISIBLE_DEVICES'] = ' 0'



def main():

    # Parameters

    BATCH_SIZE = 1

    crop_size = 48

    NUM_TESTING_STEPS = 48*50

    # Create Input and Output

    low_res_holder = tf.placeholder(tf.float32, shape=[BATCH_SIZE, crop_size, crop_size, NUM_CHENNELS])

    high_res_holder = tf.placeholder(tf.float32, shape=[BATCH_SIZE, crop_size, crop_size, NUM_CHENNELS])



    # CNN model

    low_res_image = tf.reshape(low_res_holder, shape=[-1, crop_size, crop_size, NUM_CHENNELS])

    high_res_image = tf.reshape(high_res_holder, shape=[-1, crop_size, crop_size, NUM_CHENNELS])

    inferences = SRCNN_models.create_model(MODEL_NAME, low_res_image,is_train= True)



    testing_loss = SRCNN_models.loss(inferences, high_res_image, name='testing_loss', weights_decay=0)

    srcing_loss = SRCNN_models.loss(low_res_image, high_res_image, name='src_loss', weights_decay=0)



    '''

    TensorFlow Session

    '''

    # start TensorFlow session

    init = tf.initialize_all_variables()

    saver = tf.train.Saver()

    sess = tf.InteractiveSession()

    sess.run(init)

    saver.restore(sess, 'model2/mymodel')

    mse =  np.zeros((NUM_TESTING_STEPS,1))

    mse2 =  np.zeros((NUM_TESTING_STEPS,1))

    msed = np.zeros((NUM_TESTING_STEPS,1))

    # train_epoch = multi_channel_image_generator_complex_mat("train\\ceshi\\test_Amp_3echo_FA2_for_every_picture/", patch_size_low, 1,

    #                                                         BATCH_SIZE, False,

    #                                                         seed=None)

    # train_epoch = multi_channel_image_generator_complex_mat("train\\ceshi\\test_FA2_echo2_echo_norm/", patch_size_low, 1,

    #                                                         BATCH_SIZE, False,

    #                                                         seed=None)

    # train_epoch = image_generator_mat("train\\test_Amp_all_echo_in_one_channel/", patch_size_low, 1, BATCH_SIZE, False,

    #                                   seed=None)

    # 生成队列

    # batch_low, batch_high = tfrecord_read("Amp_all_echo_one_channel.tfrecord", BATCH_SIZE, crop_size)

    batch_low, batch_high = tfrecord_read_6echo("Amp_6channel.tfrecord", BATCH_SIZE, crop_size)



    out = np.zeros((crop_size,crop_size,NUM_TESTING_STEPS))

    low = np.zeros((crop_size,crop_size,NUM_TESTING_STEPS))

    high = np.zeros((crop_size,crop_size,NUM_TESTING_STEPS))

    coord = tf.train.Coordinator()

    tf.train.start_queue_runners(coord=coord)

    for  i in range(NUM_TESTING_STEPS):

        batch_xs, batch_ys = sess.run([batch_low, batch_high])

        # tf.summary.image('input', tf.uint8(batch_xs), 10)

        recon, high_res_images, low_res_images = sess.run([inferences, high_res_image, low_res_image],

                                                          feed_dict={low_res_holder: batch_xs,

                                                                     high_res_holder: batch_ys})

        mse2[i] = sess.run(srcing_loss, feed_dict={low_res_holder: batch_xs, high_res_holder: batch_ys})

        mse[i] = sess.run(testing_loss, feed_dict={low_res_holder: batch_xs, high_res_holder: batch_ys})

        # out[:,:,i*3:3*(i+1)] = recon[0,:,:,:]

        # low[:,:,i*3:3*(i+1)] = low_res_images[0,:,:,:]

        # high[:,:,i*3:3*(i+1)] = high_res_images[0,:,:,:]

        # save for single_channel

        # out[:,:,i:(i+1)] = recon[0,:,:,:]

        # low[:,:,i:(i+1)] = low_res_images[0,:,:,:]

        # high[:,:,i:(i+1)] = high_res_images[0,:,:,:]

        # mse= 0

        # mse2[i] = src_loss

        # mse3[i] = ssiming_loss1

        # # mse4[i] = ssiming_loss2

        # recon_FA1_echo1 = recon[:, :, :, 0]

        # recon_FA1_echo2 = recon[:, :, :, 1]

        # recon_FA1_echo3 = recon[:, :, :, 2]

        # result_FA1_echo1 = recon_FA1_echo1.reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo2 = recon_FA1_echo2.reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo3 = recon_FA1_echo3.reshape(patch_size, patch_size).astype(np.float32) * 255

        # # # mse= 0

        # # # mse2[i] = src_loss

        # # # mse3[i] = ssiming_loss1

        # # # # mse4[i] = ssiming_loss2

        # recon_FA1_echo1 = recon[:,:,:,0:2]

        # # recon_FA1_echo2 = recon[:,:,:,2:4]

        # # recon_FA1_echo3 = recon[:,:,:,4:6]

        # # recon_FA2_echo1 = recon[:,:,:,6:8]

        # # recon_FA2_echo2 = recon[:,:,:,8:10]

        # # recon_FA2_echo3 = recon[:,:,:,10:12]

        # #

        # result_FA1_echo1 = np.sqrt(np.sum(np.square(recon_FA1_echo1),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo2 = np.sqrt(np.sum(np.square(recon_FA1_echo2),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo3 = np.sqrt(np.sum(np.square(recon_FA1_echo3),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo1 = np.sqrt(np.sum(np.square(recon_FA2_echo1),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo2 = np.sqrt(np.sum(np.square(recon_FA2_echo2),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo3 = np.sqrt(np.sum(np.square(recon_FA2_echo3),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # #

        # imsave('result2/final_FA1_echo1_'+str(i)+'.bmp', result_FA1_echo1)

        # imsave('result2/final_FA1_echo2_'+str(i)+'.bmp', result_FA1_echo2)

        # imsave('result2/final_FA1_echo3_'+str(i)+'.bmp',result_FA1_echo3)

        # # imsave('data/result2/final_FA1_echo1_'+str(i)+'.bmp', result_FA1_echo1)

        # # imsave('data/result2/final_FA1_echo2_'+str(i)+'.bmp', result_FA1_echo2)

        # # imsave('data/result2/final_FA1_echo3_'+str(i)+'.bmp',result_FA1_echo3)

        # # imsave('data/result/final_FA2_echo1_'+str(i)+'.bmp', result_FA2_echo1)

        # # imsave('data/result/final_FA2_echo2_'+str(i)+'.bmp', result_FA2_echo2)

        # # imsave('data/result/final_FA2_echo3_'+str(i)+'.bmp', result_FA2_echo3)

        # #

        # #

        # recon_FA1_echo1 = low_res_images[:,:,:,0]

        # recon_FA1_echo2 = low_res_images[:,:,:,1]

        # recon_FA1_echo3 = low_res_images[:,:,:,2]

        # # recon_FA1_echo1 = low_res_images[:,:,:,0:2]

        # # recon_FA1_echo2 = low_res_images[:,:,:,2:4]

        # # recon_FA1_echo3 = low_res_images[:,:,:,4:6]

        # # recon_FA2_echo1 = low_res_images[:,:,:,6:8]

        # # recon_FA2_echo2 = low_res_images[:,:,:,8:10]

        # # recon_FA2_echo3 = low_res_images[:,:,:,10:12]

        # #

        # result_FA1_echo1 = recon_FA1_echo1.reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo2 = recon_FA1_echo2.reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo3 = recon_FA1_echo3.reshape(patch_size, patch_size).astype(np.float32) * 255

        # # recon_FA1_echo1 = low_res_images[:,:,:,0:2]

        # # recon_FA1_echo2 = low_res_images[:,:,:,2:4]

        # # recon_FA1_echo3 = low_res_images[:,:,:,4:6]

        # # recon_FA2_echo1 = low_res_images[:,:,:,6:8]

        # # recon_FA2_echo2 = low_res_images[:,:,:,8:10]

        # # recon_FA2_echo3 = low_res_images[:,:,:,10:12]

        # #

        # result_FA1_echo1 = np.sqrt(np.sum(np.square(recon_FA1_echo1),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo2 = np.sqrt(np.sum(np.square(recon_FA1_echo2),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo3 = np.sqrt(np.sum(np.square(recon_FA1_echo3),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo1 = np.sqrt(np.sum(np.square(recon_FA2_echo1),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo2 = np.sqrt(np.sum(np.square(recon_FA2_echo2),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo3 = np.sqrt(np.sum(np.square(recon_FA2_echo3),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # #

        # imsave('result2/low_FA1_echo1_' + str(i) + '.bmp', result_FA1_echo1)

        # imsave('result2/low_FA1_echo2_' + str(i) + '.bmp', result_FA1_echo2)

        # imsave('result2/low_FA1_echo3_' + str(i) + '.bmp', result_FA1_echo3)

        # # imsave('data/result2/low_FA1_echo1_'+str(i)+'.bmp', result_FA1_echo1)

        # # imsave('data/result2/low_FA1_echo2_'+str(i)+'.bmp', result_FA1_echo2)

        # # imsave('data/result2/low_FA1_echo3_'+str(i)+'.bmp',result_FA1_echo3)

        # # imsave('data/result/low_FA2_echo1_' + str(i) + '.bmp', result_FA2_echo1)

        # # imsave('data/result/low_FA2_echo2_' + str(i) + '.bmp', result_FA2_echo2)

        # # imsave('data/result/low_FA2_echo3_' + str(i) + '.bmp', result_FA2_echo3)

        # #

        # recon_FA1_echo1 = high_res_images[:, :, :, 0]

        # recon_FA1_echo2 = high_res_images[:, :, :, 1]

        # recon_FA1_echo3 = high_res_images[:, :, :, 2]

        # # recon_FA1_echo1 = high_res_images[:,:,:,:]

        # # recon_FA1_echo2 = high_res_images[:,:,:,2:4]

        # # recon_FA1_echo3 = high_res_images[:,:,:,4:6]

        # # recon_FA2_echo1 = high_res_images[:,:,:,6:8]

        # # recon_FA2_echo2 = high_res_images[:,:,:,8:10]

        # # recon_FA2_echo3 = high_res_images[:,:,:,10:12]

        # result_FA1_echo1 = recon_FA1_echo1.reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo2 = recon_FA1_echo2.reshape(patch_size, patch_size).astype(np.float32) * 255

        # result_FA1_echo3 = recon_FA1_echo3.reshape(patch_size, patch_size).astype(np.float32) * 255

        # # # recon_FA1_echo1 = high_res_images[:,:,:,0:2]

        # # recon_FA1_echo2 = high_res_images[:,:,:,2:4]

        # # recon_FA1_echo3 = high_res_images[:,:,:,4:6]

        # # recon_FA2_echo1 = high_res_images[:,:,:,6:8]

        # # recon_FA2_echo2 = high_res_images[:,:,:,8:10]

        # # recon_FA2_echo3 = high_res_images[:,:,:,10:12]

        # # result_FA1_echo1 = np.sqrt(np.sum(np.square(recon_FA1_echo1),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA1_echo2 = np.sqrt(np.sum(np.square(recon_FA1_echo2),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA1_echo3 = np.sqrt(np.sum(np.square(recon_FA1_echo3),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo1 = np.sqrt(np.sum(np.square(recon_FA2_echo1),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo2 = np.sqrt(np.sum(np.square(recon_FA2_echo2),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # # result_FA2_echo3 = np.sqrt(np.sum(np.square(recon_FA2_echo3),axis =3)).reshape(patch_size, patch_size).astype(np.float32) * 255

        # #

        # imsave('result2/high_FA1_echo1_' + str(i) + '.bmp', result_FA1_echo1)

        # imsave('result2/high_FA1_echo2_' + str(i) + '.bmp', result_FA1_echo2)

        # imsave('result2/high_FA1_echo3_' + str(i) + '.bmp', result_FA1_echo3)

        # imsave('data/result2/high_FA1_echo1_'+str(i)+'.bmp', result_FA1_echo1)

        # imsave('data/result2/high_FA1_echo2_'+str(i)+'.bmp', result_FA1_echo2)

        # imsave('data/result2/high_FA1_echo3_'+str(i)+'.bmp',result_FA1_echo3)

        # imsave('data/result/high_FA2_echo1_' + str(i) + '.bmp', result_FA2_echo1)

        # imsave('data/result/high_FA2_echo2_' + str(i) + '.bmp', result_FA2_echo2)

        # imsave('data/result/high_FA2_echo3_' + str(i) + '.bmp', result_FA2_echo3)

        print('i:%d, test_MSE: %.7f, src_MSE: %.7f' % (i,mse[i], mse2[i]))

    # save for single_channel

    # sio.savemat('out.mat', {'out': out})

    # sio.savemat('low.mat', {'low': low})

    # sio.savemat('high.mat', {'high': high})

    # mse /= NUM_TESTING_STEPS

    # mse2 /= 192

    print('i: %d ,min_test_MSE: %.7f, max_test_MSE: %.7f, ave_test_MSE: %.7f, min_src_MSE: %.7f, max_src_MSE: %.7f, ave_src_MSE: %.7f' % (i, mse.min(),mse.max(),mse.mean(), mse2.min(),mse2.max(),mse2.mean()))



if __name__ == '__main__':

    main()
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化