# var_values = np.hstack((np.arange(2, 20, 2), np.arange(20, 44, 4)))
# var_values = np.arange(0, 104, 4)
# var_values = np.hstack((np.arange(0, 100, 2), np.arange(100, 700, 10),
# np.arange(700, 780, 5), np.arange(780, 784, 2)))
select_way = 'SequencedMeasure' # 'SequencedMeasure', 'RandomMeasure', 'MaxSEE', 'Variance'
generate_way = 'eigs' # 'eigs': grey images, 'rand': black-and-white (Lei's way)
# Initialize parameters
dataset = 'fashion_mnist'
samples = 'all'
num_features = 100
which_class = 3
chi = 40
# =============================================================
para = parameters_gtn_one_class()
para['dct'] = False
para['d'] = 2
para['step'] = 0.2 # initial gradient step
para['if_save'] = True
para['if_load'] = True
para['dataset'] = dataset
para['class'] = which_class
para['chi'] = chi
psnr_av = np.zeros(var_values.shape)
mse_av = np.zeros(var_values.shape)
ssim = np.zeros(var_values.shape)
b = TNmachineLearning.MachineLearningFeatureMap(para['d'], para['dataset'])
if var_name is not 'which_class':
b.load_data(data_path=os.path.join(b.project_path, '..\\..\\MNIST\\' + para['dataset'] + '\\'),
def quantum_jpeg(para_tot=None):
if para_tot is None:
para_tot = parameters_qjpg()
print('Preparing image')
exp = expression_save(para_tot)
exp = os.path.join(para_tot['data_path'], exp)
if os.path.isfile(exp) and para_tot['if_load']:
a, generator = load_pr(exp, ['a', 'mps'])
else:
a = qjpg(file=para_tot['file'],
b_size=para_tot['block_size'],
is_grey=True)
a.cut2blocks()
a.dct_blocks()
a.pre_process_data_before_ml(which=2)
para = parameters_gtn_one_class()
para['chi'] = para_tot['chi']
para['dataset'] = 'custom'
para['if_save'] = False
para['if_load'] = False
para['dct'] = False
generator = None
if 'real' in para_tot['tasks']:
print('Train in the real space')
for n in range(para_tot['reorder_time']):
generator = gtn_one_class(para=para,
images=a.blocks.squeeze())[0]
if n != (para_tot['reorder_time'] - 1):
order = generator.mps.markov_measurement(
if_restore=False)[0]
# ent = generator.mps.calculate_single_entropy()[0]
# order = np.argsort(ent)[::-1]
a.reorder_features(order, which=0)
if 'freq' in para_tot['tasks']:
for n in range(para_tot['reorder_time']):
print('Train in the frequency space： reorder time = ' + str(n))
generator = gtn_one_class(para=para,
images=a.blocks_dct.squeeze())[0]
if n != (para_tot['reorder_time'] - 1):
order = generator.mps.markov_measurement(
if_restore=False)[0]
# ent = generator.mps.calculate_single_entropy()[0]
# order = np.argsort(ent)[::-1]
a.reorder_features(order, which=1)
save_pr(para_tot['data_path'], exp, [a, generator, para_tot],
['a', 'mps', 'para_tot'])
# a.show_image()
if 'recover' in para_tot['tasks']:
blocks = a.encode_with_cutoff_dct(para_tot['pixel_cutoff'])
image_gtn1 = decode_with_generative_mps(blocks, generator)
image_cutoff = decode_with_cutoff_dct(blocks)
blocks_jpg = a.encode_standard_jpeg(para_tot['pixel_cutoff'])
image_jpg = decode_jpeg(blocks_jpg)
io.imsave('../data_QJPG/before.jpg', a.image.squeeze())
io.imsave('../data_QJPG/0cut.jpg', image_cutoff.squeeze())
io.imsave('../data_QJPG/1JPGway.jpg', image_jpg.squeeze())
io.imsave('../data_QJPG/2GTNway.jpg', image_gtn1.squeeze())
p1 = psnr(a.image0, image_jpg)
p2 = psnr(a.image0, image_cutoff)
p3 = psnr(a.image0, image_gtn1)
print('The PSNRs for jpg, cut-off, and gtn are %g, %g, and %g' %
(p1, p2, p3))
