def run_test(self):
test_data, test_label = test_input_setup(self)
print("Testing...")
start_time = time.time()
result = np.clip(
self.pred.eval({
self.images: test_data,
self.labels: test_label,
self.batch: 1
}), 0, 1)
passed = time.time() - start_time
img1 = tf.convert_to_tensor(test_label, dtype=tf.float32)
img2 = tf.convert_to_tensor(result, dtype=tf.float32)
psnr = self.sess.run(tf.image.psnr(img1, img2, 1))
ssim = self.sess.run(tf.image.ssim(img1, img2, 1))
print("Took %.3f seconds, PSNR: %.6f, SSIM: %.6f" %
(passed, psnr, ssim))
result = merge(self, result)
image_path = os.path.join(os.getcwd(), self.output_dir)
image_path = os.path.join(image_path, "test_image.png")
array_image_save(result, image_path)
def run_test(self):
nx, ny = test_input_setup(self)
data_dir = os.path.join('{}'.format(self.checkpoint_dir), "test.h5")
test_data, test_label = read_data(data_dir)
test_image = merge(test_data, [nx, ny]).squeeze()
print("nx =", nx, "\nny =", ny)
print(test_image.shape)
print(test_data.shape)
# print(test_data)
print("Testing...")
start_time = time.time()
result = self.pred.eval({
self.images: test_data,
self.labels: test_label,
self.batch: nx * ny
})
print("Took %.3f seconds" % (time.time() - start_time))
result = merge(result, [nx, ny])
result = result.squeeze()
image_path = os.path.join(os.getcwd(), self.output_dir)
# image_path = os.path.join(image_path, "test_image.png")
image_path = os.path.join(image_path, "test_image.png")
array_image_save(result * 255, image_path)
def run_test(self):
print("[INFO] Testing . . .")
# NOTE only outputting 15 test examples
for k in range(0, 15):
np.random.seed(k)
test_data, test_label = test_input_setup(self, np.random.randint(0,200))
print("Testing...")
start_time = time.time()
result = np.clip(self.pred.eval({self.images: test_data, self.labels: test_label, self.batch: 1}), 0, 1)
print(" RESULT ", result.shape)
print(" test_data ", test_data.shape)
print(" test_label ", test_label.shape)
passed = time.time() - start_time
img1 = tf.convert_to_tensor(test_label, dtype=tf.float32)
img2 = tf.convert_to_tensor(result, dtype=tf.float32)
psnr = self.sess.run(tf.image.psnr(img1, img2, 1))
ssim = self.sess.run(tf.image.ssim(img1, img2, 1))
print("Took %.3f seconds, PSNR: %.6f, SSIM: %.6f" % (passed, psnr, ssim))
image_path = os.path.join(os.getcwd(), self.output_dir)
image_path = os.path.join(image_path, "test_image.png")
import matplotlib.pyplot as plt
fig = plt.figure(figsize=(11,11))
ax1 = fig.add_subplot(131); ax1.imshow(test_data.reshape(128,128)); ax1.axis('off'); ax1.set_title('LR')
ax2 = fig.add_subplot(132); ax2.imshow(test_label.reshape(384,384)); ax2.axis('off'); ax2.set_title('HR')
ax3 = fig.add_subplot(133); ax3.imshow(result.reshape(384,384)); ax3.axis('off'); ax3.set_title('Predicted')
plt.savefig('./result/' + str(k) + '_test.png')
# Calculate Test MSE
all = mse_models(self)
mse_all = []
for img in all:
test_data = img[0]
test_label = img[1]
result = np.clip(self.pred.eval({self.images: test_data, self.labels: test_label, self.batch: 1}), 0, 1)
result = result.reshape(384,384)
test_label = test_label.reshape(384,384)
mse = np.sum((np.abs(result.astype("float") - test_label.astype("float"))) ** 2)
mse /= float(result.shape[0] * result.shape[1])
mse_all.append(mse)
print("##########################")
print("MSE IS: ", np.mean(mse_all))
def run_test(self):
test_data, test_label, nx, ny = test_input_setup(self)
print("Testing...")
start_time = time.time()
result = self.pred.eval({self.images: test_data, self.labels: test_label, self.batch: nx * ny})
print("Took %.3f seconds" % (time.time() - start_time))
result = merge(result, [nx, ny, self.c_dim])
result = result.squeeze()
image_path = os.path.join(os.getcwd(), self.output_dir)
image_path = os.path.join(image_path, "test_image.png")
array_image_save(result * 255, image_path)
def run_test(self):
nx, ny = test_input_setup(self)
data_dir = os.path.join('./{}'.format(self.checkpoint_dir), 'test.h5')
test_data, test_label = read_data(data_dir)
print(
'................................Testing..................................'
)
result = self.pred.eval({
self.images: test_data,
self.labels: test_label,
self.batch: nx * ny
})
result = merge(result, [nx, ny])
result = result.squeeze()
image_path = os.path.join(os.getcwd(), self.output_dir)
image_path = os.path.join(image_path, "test_image.png")
array_image_save(result * 255, image_path)
def run_test(self):
# 在test_input_setup中提取出image的Y通道数据下采样后归一化，进行reshape和padding=4
test_data_label_arr = test_input_setup(self)
print("Testing...")
i = 0
for test_data, test_label in test_data_label_arr:
start_time = time.time()
result = np.clip(self.pred.eval({self.images: test_data, self.labels: test_label, self.batch: 1}), 0, 1)
passed = time.time() - start_time
img1 = tf.convert_to_tensor(test_label, dtype=tf.float32)
img2 = tf.convert_to_tensor(result, dtype=tf.float32)
psnr = self.sess.run(tf.image.psnr(img1, img2, 1))
ssim = self.sess.run(tf.image.ssim(img1, img2, 1))
print("Took %.3f seconds, PSNR: %.6f, SSIM: %.6f" % (passed, psnr, ssim))
# 将Y通道处理后合并到原图的CbCr
result = merge(self, result, i)
image_path = os.path.join(os.getcwd(), self.output_dir)
image_path = os.path.join(image_path, "test_image_" + str(i) + ".png")
i = i + 1
array_image_save(result, image_path)
def run_test(self):
# 读取测试数据集图片
data_dir = os.path.join(os.sep,
(os.path.join(os.getcwd(), self.TestData_dir)),
"Set5")
data = sorted(glob.glob(os.path.join(data_dir, "*.bmp")))
data_img = prepare_data(self, dataset=self.TestData_dir)
# 生成列表记录每次的psnr，准备用于psnr的平均值求取
all_psnr_bic, all_psnr_fsr = [], []
# 遍历路径下的图片，并进行测试
for i in range(len(data)):
# 制作测试用的h5文件，并读取测试图和各类输入图
nx, ny, im_y_hr_same, im_y_bic_same, im_cb_bic_same, im_cr_bic_same, color_im = test_input_setup(
self, i, data_img)
data_dir = os.path.join('./{}'.format(self.checkpoint_dir),
"test.h5")
test_data, test_label = read_data(data_dir)
# 开始测试
print("Testing...")
start_time = time.time()
result = self.SRpred.eval({
self.LRs: test_data,
self.HRs: test_label,
self.batch: nx * ny
})
print("Took %.3f seconds" % (time.time() - start_time))
# 对输出结果图片转换size储存形式
result = merge(result, [nx, ny])
result = result.squeeze() # 从数组的形状中删除单维度条目，即把shape中为1的维度去掉
# 对输出结果、bicubic、原图去除边界，提取干净数据
height, weight = result.shape[0], result.shape[1]
result = result[2:height - 2, 2:weight - 2]
# result = result[1:height-3, 1:weight-3] # matlab官方在*3上的做法，但此处效果不对，应该是卷积核的区别
im_y_hr_same = im_y_hr_same[2:height - 2, 2:weight - 2]
im_y_bic_same = im_y_bic_same[2:height - 2, 2:weight - 2]
# 对输出图片进行批量格式化命名
dst00 = os.path.join(os.path.abspath(self.output_dir),
'00' + format(str(i), '0>3s') + '_BIC.bmp')
dst10 = os.path.join(os.path.abspath(self.output_dir),
'00' + format(str(i), '0>3s') + '_FSR.bmp')
dst20 = os.path.join(os.path.abspath(self.output_dir),
'00' + format(str(i), '0>3s') + '_HR.bmp')
# 保存各类测试结果图
array_image_save(im_y_bic_same * 255, dst00,
color_im) # 保存bicubic输出灰度图
array_image_save(im_y_hr_same * 255, dst20, color_im) # 保存hr输出灰度图
array_image_save(result * 255, dst10, color_im) # 保存fsrcnn输出灰度图
# 计算每次bicubic和fsrcnn测试结果的psnr
psnr_bic = self.psnr_computer(im_y_hr_same, im_y_bic_same)
psnr_fsr = self.psnr_computer(im_y_hr_same, result)
all_psnr_bic.append(psnr_bic)
all_psnr_fsr.append(psnr_fsr)
print('PSNR for Bicubic:', psnr_bic, 'dB\n'
'PSNR for FSRCNN:', psnr_fsr, 'dB\n')
if color_im == 1:
# 对cb,cr通道去除边界，提取干净数据
im_cb_bic_same = im_cb_bic_same[2:height - 2, 2:weight - 2]
im_cr_bic_same = im_cr_bic_same[2:height - 2, 2:weight - 2]
# 将y，cb，cr通道转为rgb通道，并对溢出值进行处理
bic_color = color_image_save(im_y_bic_same, im_cb_bic_same,
im_cr_bic_same)
fsr_color = color_image_save(result, im_cb_bic_same,
im_cr_bic_same)
hr_color = color_image_save(im_y_hr_same, im_cb_bic_same,
im_cr_bic_same)
# 对输出图片进行批量格式化命名
dst01 = os.path.join(
os.path.abspath(self.output_dir),
'00' + format(str(i), '0>3s') + '_BIC_color.bmp')
dst11 = os.path.join(
os.path.abspath(self.output_dir),
'00' + format(str(i), '0>3s') + '_FSR_color.bmp')
dst21 = os.path.join(
os.path.abspath(self.output_dir),
'00' + format(str(i), '0>3s') + '_HR_color.bmp')
# 保存各类测试结果图
skio.imsave(dst01, bic_color) # 保存bicubic输出彩色图
skio.imsave(dst11, fsr_color) # 保存hr输出彩色图
skio.imsave(dst21, hr_color) # 保存fsrcnn输出彩色图
# 计算总的bicubic和fsrcnn测试结果的平均psnr
mean_psnr_bic, mean_psnr_fsr = np.mean(all_psnr_bic), np.mean(
all_psnr_fsr)
print('Mean PSNR for Bicubic:', mean_psnr_bic, 'dB\n'
'Mean PSNR for FSRCNN:', mean_psnr_fsr, 'dB\n')