def tensor_gray_4D_to_image(tensor,do_colorize = False,do_scale=False,force_rows_dim3=None,force_rows_dim2=None):
sub_image = tensor_gray_3D_to_image(tensor[:, :, :, 0],force_rows_dim2)
h, w = sub_image.shape[0], sub_image.shape[1]
if force_rows_dim3 is None:
R = tools_image.numerical_devisor(tensor.shape[3])
else:
R = force_rows_dim3
C = int(tensor.shape[3]/R)
if do_colorize:
image = numpy.zeros((h * R, w * C, 3), dtype=numpy.float32)
else:
image = numpy.zeros((h * R, w * C), dtype=numpy.float32)
for i in range (0,tensor.shape[3]):
col, row = i % C, int(i / C)
sub_image = tensor_gray_3D_to_image(tensor[:,:,:,i])
if do_colorize:
sub_image = tools_image.saturate(sub_image)
apply_blue_shift(sub_image,(col+row)%2)
if do_colorize:
image[h * row:h * row + h, w * col:w * col + w,:] = sub_image[:, :, :]
else:
image[h * row:h * row + h, w * col:w * col + w] = sub_image[:, :]
if do_scale==True:
image -= image.min()
image *= 255 / image.max()
return image
def tensor_gray_3D_to_image(tensor,
do_colorize=False,
do_scale=False,
force_rows_dim2=None):
if force_rows_dim2 is None:
rows = tools_image.numerical_devisor(tensor.shape[2])
else:
rows = force_rows_dim2
h, w, R, C = tensor.shape[0], tensor.shape[1], rows, int(tensor.shape[2] /
rows)
image = numpy.zeros((h * R, w * C), dtype=numpy.float32)
for i in range(0, tensor.shape[2]):
col, row = i % C, int(i / C)
image[h * row:h * row + h, w * col:w * col + w] = tensor[:, :, i]
if do_colorize:
image = colorize_chess(image, tensor.shape[0], tensor.shape[1])
if do_scale == True:
image -= image.min()
image *= 255 / image.max()
return image.astype(dtype=numpy.uint8)
def image_to_tensor_color_4D(image, shape):
tensor = numpy.zeros(shape, numpy.float32)
h, w = shape[0], shape[1]
rows = tools_image.numerical_devisor(shape[3])
C = int(tensor.shape[3] / rows)
for i in range(0, 96):
col, row = i % C, int(i / C)
tensor[:, :, :, i] = image[h * row:h * row + h, w * col:w * col + w]
return tensor
def tensor_color_4D_to_image(tensor):
rows = tools_image.numerical_devisor(tensor.shape[3])
h, w, R, C = tensor.shape[0], tensor.shape[1], rows, int(tensor.shape[3] /
rows)
image = numpy.zeros((h * R, w * C, tensor.shape[2]), dtype=numpy.float32)
for i in range(0, tensor.shape[3]):
col, row = i % C, int(i / C)
image[h * row:h * row + h, w * col:w * col + w, :] = tensor[:, :, :, i]
return image
def tensor_gray_1D_to_image(tensor,orientation = 'landscape'):
rows = tools_image.numerical_devisor(tensor.shape[0])
cols = int(tensor.shape[0] / rows)
if orientation=='landscape':
image = numpy.reshape(tensor,(numpy.minimum(rows,cols),numpy.maximum(rows, cols)))
else:
image = numpy.reshape(tensor, (numpy.maximum(rows, cols), numpy.minimum(rows, cols)))
return image
def tensor_gray_3D_to_image(tensor, do_colorize=False):
rows = tools_image.numerical_devisor(tensor.shape[2])
h, w, R, C = tensor.shape[0], tensor.shape[1], rows, int(tensor.shape[2] /
rows)
image = numpy.zeros((h * R, w * C), dtype=numpy.float32)
for i in range(0, tensor.shape[2]):
col, row = i % C, int(i / C)
image[h * row:h * row + h, w * col:w * col + w] = tensor[:, :, i]
if do_colorize:
image = colorize_chess(image, tensor.shape[0], tensor.shape[1])
return image
def init_model(self, input_dim, out_dim, grayscale):
self.model = Sequential()
rows = tools_image.numerical_devisor(int(input_dim))
cols = int(input_dim / rows)
self.model.add(Reshape((rows, cols, 1), input_shape=[input_dim]))
self.model.add(Conv2D(32, (2, 2), activation='linear', padding='same'))
self.model.add(UpSampling2D((2, 2)))
self.model.add(Flatten())
self.model.add(Dense((out_dim), activation='linear'))
self.model.compile(optimizer='adam',
loss='mean_squared_error',
metrics=['accuracy'])
return
def tensor_color_4D_to_image(tensor,do_chess=False,force_rows_dim3=None,force_rows_dim2=None):
if force_rows_dim2 is None:
rows = tools_image.numerical_devisor(tensor.shape[3])
else:
rows = force_rows_dim2
h, w, R, C = tensor.shape[0], tensor.shape[1], rows, int(tensor.shape[3] / rows)
image = numpy.zeros((h * R,w * C, tensor.shape[2]),dtype=numpy.float32)
for i in range(0, tensor.shape[3]):
col, row = i % C, int(i / C)
sub_image = tensor[:, :, :, i]
if do_chess:
apply_blue_shift(sub_image,(col+row)%2)
image[h * row:h * row + h, w * col:w * col + w, :] = sub_image
return image
def tensor_gray_4D_to_image(tensor, do_colorize=False):
sub_image = tensor_gray_3D_to_image(tensor[:, :, :, 0])
h, w = sub_image.shape[0], sub_image.shape[1]
R = tools_image.numerical_devisor(tensor.shape[3])
C = int(tensor.shape[3] / R)
if do_colorize:
image = numpy.zeros((h * R, w * C, 3), dtype=numpy.float32)
else:
image = numpy.zeros((h * R, w * C), dtype=numpy.float32)
for i in range(0, tensor.shape[3]):
col, row = i % C, int(i / C)
sub_image = tensor_gray_3D_to_image(tensor[:, :, :, i], do_colorize)
if do_colorize:
image[h * row:h * row + h,
w * col:w * col + w, :] = sub_image[:, :, :]
else:
image[h * row:h * row + h, w * col:w * col + w] = sub_image[:, :]
return image
