def De_test(myDeconvNN_CS, with_new_csphi, myflag):
test_data_list = "./Training_Data_L/validation1/"
testImgList = tdp.eachFile(test_data_list)
testImgArrayList_ori = tdp.crop_img(testImgList,
crop_strides=tdp.crops_width)
testImgArrayList = tdp.up_dim(testImgArrayList_ori)
#myflag.model = "test"
#myflag.check_batch_size()
#print("In test, myflag.batch_size:", myflag.batch_size)
CS_Phi = ge_csphi(with_new_csphi)
# test数据集的总损失
total_test_cost = 0.
batch_num = int(len(testImgArrayList) / myflag.batch_size)
#print("testImgArrayList[0].shape:", testImgArrayList[0].shape)
for i in range(batch_num):
img_batch = get_block_from_data(testImgArrayList, myflag.batch_size, i)
test_cost_batch = myDeconvNN_CS.calc_cost(X_2D=img_batch,
CS_Phi=CS_Phi)
total_test_cost += test_cost_batch
print("In test phase, batch_size:", myflag.batch_size)
#print("De_test_batch_cost=", "{:.9f}".format(total_test_cost / batch_num))
print("De_one_test_cost=",
"{:.9f}".format(total_test_cost / (batch_num * myflag.batch_size)))
def cal_each_mean_PSNR(oriImgDir, reconImgDir):
oriImgFullPaths = tdp.eachFile(oriImgDir)
reconImgFullPaths = tdp.eachFile(reconImgDir)
#oriImgList = []
#reconImgList = []
oridict = {}
redict = {}
for oriImgFullPath in oriImgFullPaths:
oriImgName = os.path.split(oriImgFullPath)[1].split(".")[0]
oridict[oriImgName] = oriImgFullPath
'''
a.append(os.path.split(oriImgFullPath)[1])
oriImg = Image.open(oriImgFullPath)
oriImgList.append(oriImg)
'''
for reconImgFullPath in reconImgFullPaths:
reconImgName = os.path.split(reconImgFullPath)[1].split('.')[0]
redict[reconImgName] = reconImgFullPath
'''
b.append(os.path.split(reconImgFullPath)[1])
reconImg = Image.open(reconImgFullPath)
reconImgList.append(reconImg)
'''
'''
print("ori:", a)
print("re:", b)
'''
PSNR_dict = {}
totalPSNR = 0
#for i in range(len(oriImgList)):
for i in oridict:
#print(i)
oriImg = Image.open(oridict[i])
reImg = Image.open(redict[i])
somePSNR = sR.calc_PSNR(generate_img=reImg, original_img=oriImg)
#somePSNR = sR.calc_PSNR(generate_img=reconImgList[i], original_img=oriImgList[i])
#PSNR_dict[os.path.split(oriImgFullPaths[i])[1]] = somePSNR
PSNR_dict[i] = somePSNR
totalPSNR += somePSNR
#meanPSNR = totalPSNR/len(oriImgList)
meanPSNR = totalPSNR / len(oridict)
print("PSNR of each pair:", PSNR_dict)
print("meanPSNR:", meanPSNR)
#print("without flinstone, meanPSNR:", (totalPSNR-PSNR_dict['flinstones'])/(len(oridict)-1))
return
def show_from_De1():
Decnn = De1.myDeconvNN_CS
dirs = "./Training_Data_L/Test/Set1/"
imgList = tdp.eachFile(dirs)
img = Image.open(imgList[0])
img.show()
imgArrayList = tdp.crop_img(imgList)
imgArrayList_up = tdp.up_dim(imgArrayList)
print("len(imgArrayList):", len(imgArrayList))
result_img = get_Result_img(imgArrayList=imgArrayList_up,
original_img=img,
nn=Decnn)
result_img.show()
PSNR = calc_PSNR(result_img, img)
print("PSNR:", PSNR)
def save_reconImg_from_nn(nn):
dirs = "./Training_Data_L/test_images/noiseless/"
saveDir = './Training_Data_L/re_test_images/noiseless/v1/mr_0_01_e1000/'
imgList = tdp.eachFile(dirs)
#print("last one:", imgList[len(imgList)-1])
crop_strides = tdp.crops_width
imgArrayList = tdp.crop_img(imgList, crop_strides=crop_strides)
imgArrayList_up = tdp.up_dim(imgArrayList)
start = time.clock()
save_Result_img(imgArrayList=imgArrayList_up,
oriFileDir=dirs,
saveDir=saveDir,
nn=nn,
crop_strides=crop_strides,
start=start)
return
def show_img_from_nn(nn):
dirs = "./Training_Data_L/Test/Set1/"
imgList = tdp.eachFile(dirs)
img = Image.open(imgList[0])
img.show()
crop_strides = tdp.crops_width
imgArrayList = tdp.crop_img(imgList, crop_strides=crop_strides)
imgArrayList_up = tdp.up_dim(imgArrayList)
result_img = get_Result_img(imgArrayList=imgArrayList_up,
original_img=img,
nn=nn,
crop_strides=crop_strides)
result_img.show()
#denoised_img_array = BM3D_denoise.ge_final(result_img)
#print("enoised_img_array.shape:", denoised_img_array.shape)
#print("denoised_img_array:", denoised_img_array)
#denoised_img = Image.fromarray(denoised_img_array)
#print("show the denoised image.")
#denoised_img.show()
PSNR_mine1 = calc_PSNR(result_img, img)
#PSNR_mine2 = calc_PSNR(denoised_img, img)
print("PSNR_mine1 between nondenoised_img and ori_img:", PSNR_mine1)
def De_train(myDeconvNN_CS, with_new_csphi, myflag):
train_data_dirs = "./Training_Data_L/Train/"
trainImgList = tdp.eachFile(train_data_dirs)
trainImgArrayList_ori = tdp.crop_img(trainImgList)
trainImgArrayList = tdp.up_dim(trainImgArrayList_ori)
myflag.model = "train"
myflag.check_batch_size()
CS_Phi = ge_csphi(with_new_csphi)
# train中平均每个batch的总损失
batch_num = int(len(trainImgArrayList) / myflag.batch_size)
for epoch in range(training_epochs):
total_train_cost = 0.
for i in range(batch_num):
image_batch = get_block_from_data(trainImgArrayList,
myflag.batch_size, i)
train_cost_batch = myDeconvNN_CS.partial_fit(X_2D=image_batch,
CS_Phi=CS_Phi)
total_train_cost += train_cost_batch
if epoch % display_step == 0:
print(
"Epoch:", '%04d' % (epoch + 1), "De_one_train_cost=",
"{:.9f}".format(total_train_cost /
(myflag.batch_size * batch_num)))
'''
print("Epoch:", '%04d' % (epoch + 1), "De_train_batch_cost=",
"{:.9f}".format((total_train_cost / batch_num)))
'''
if (epoch + 1) % 10 == 0:
De_test(myDeconvNN_CS=myDeconvNN_CS,
with_new_csphi=False,
myflag=myflag)
print("In train phase, batch_size:", myflag.batch_size)
print("De_one_train_cost=",
"{:.9f}".format(total_train_cost / (myflag.batch_size * batch_num)))