This example demonstrates the use of onlinecdl.OnlineConvBPDNDictLearn for learning a convolutional dictionary from a set of training images. The dictionary is learned using the online dictionary learning algorithm proposed in [21]. This variant of the example uses the GPU accelerated version of onlinecdl within the sporco.cupy subpackage.
from __future__ import print_function
from builtins import input
from builtins import range
import numpy as np
from sporco import util
from sporco import plot
plot.config_notebook_plotting()
from sporco.cupy import cupy_enabled, np2cp, cp2np
from sporco.cupy import select_device_by_load, gpu_info
from sporco.cupy import cp
from sporco.cupy.dictlrn import onlinecdl
Load training images.
exim = util.ExampleImages(scaled=True, zoom=0.25)
S1 = exim.image('barbara.png', idxexp=np.s_[10:522, 100:612])
S2 = exim.image('kodim23.png', idxexp=np.s_[:, 60:572])
S3 = exim.image('monarch.png', idxexp=np.s_[:, 160:672])
S4 = exim.image('sail.png', idxexp=np.s_[:, 210:722])
S5 = exim.image('tulips.png', idxexp=np.s_[:, 30:542])
S = np.stack((S1, S2, S3, S4, S5), axis=3)
Highpass filter training images.
npd = 16
fltlmbd = 5
sl, sh = util.tikhonov_filter(S, fltlmbd, npd)
Construct initial dictionary.
np.random.seed(12345)
D0 = np.random.randn(8, 8, 3, 64)
Set regularization parameter and options for dictionary learning solver.
lmbda = 0.2
opt = onlinecdl.OnlineConvBPDNDictLearn.Options({
'Verbose': True, 'ZeroMean': False, 'eta_a': 10.0,
'eta_b': 20.0, 'DataType': np.float32,
'CBPDN': {'rho': 5.0, 'AutoRho': {'Enabled': True},
'RelaxParam': 1.8, 'RelStopTol': 1e-7, 'MaxMainIter': 50,
'FastSolve': False, 'DataType': np.float32}})
Create solver object and solve.
if not cupy_enabled():
print('CuPy/GPU device not available: running without GPU acceleration\n')
else:
id = select_device_by_load()
info = gpu_info()
if info:
print('Running on GPU %d (%s)\n' % (id, info[id].name))
d = onlinecdl.OnlineConvBPDNDictLearn(np2cp(D0), lmbda, opt)
iter = 50
d.display_start()
for it in range(iter):
img_index = np.random.randint(0, sh.shape[-1])
d.solve(np2cp(sh[..., [img_index]]))
d.display_end()
D1 = cp2np(d.getdict())
print("OnlineConvBPDNDictLearn solve time: %.2fs" % d.timer.elapsed('solve'))
Running on GPU 1 (Tesla K40c) Itn X r X s X ρ D cnstr D dlt D η ---------------------------------------------------------------- 0 9.81e-04 1.58e-03 5.00e+00 8.04e+01 6.14e+00 5.00e-01 1 1.86e-03 1.68e-03 5.00e+00 7.55e+01 4.37e+00 4.76e-01 2 2.86e-03 1.96e-03 5.00e+00 2.51e+01 2.65e+00 4.55e-01 3 1.79e-03 1.90e-03 5.00e+00 5.09e+01 2.28e+00 4.35e-01 4 2.37e-03 1.71e-03 5.00e+00 2.09e+01 1.65e+00 4.17e-01 5 1.80e-03 1.47e-03 5.00e+00 3.52e+01 1.91e+00 4.00e-01 6 2.27e-03 3.29e-03 5.00e+00 3.57e+01 2.19e+00 3.85e-01 7 1.69e-03 1.95e-03 5.00e+00 4.50e+01 2.16e+00 3.70e-01 8 1.65e-03 1.44e-03 5.00e+00 3.18e+01 1.71e+00 3.57e-01 9 2.02e-03 3.08e-03 5.00e+00 3.20e+01 1.82e+00 3.45e-01 10 2.09e-03 1.75e-03 5.00e+00 1.72e+01 1.47e+00 3.33e-01 11 1.86e-03 2.87e-03 5.00e+00 3.01e+01 1.64e+00 3.23e-01 12 2.09e-03 1.87e-03 5.00e+00 2.56e+01 1.82e+00 3.12e-01 13 2.40e-03 2.09e-03 5.00e+00 1.53e+01 1.20e+00 3.03e-01 14 1.90e-03 2.27e-03 5.00e+00 3.59e+01 1.87e+00 2.94e-01 15 1.97e-03 1.71e-03 5.00e+00 1.44e+01 1.04e+00 2.86e-01 16 2.13e-03 2.05e-03 5.00e+00 2.23e+01 1.40e+00 2.78e-01 17 2.17e-03 3.41e-03 5.00e+00 2.64e+01 1.79e+00 2.70e-01 18 2.04e-03 1.87e-03 5.00e+00 1.34e+01 1.03e+00 2.63e-01 19 1.79e-03 2.17e-03 5.00e+00 3.07e+01 1.50e+00 2.56e-01 20 1.98e-03 1.94e-03 5.00e+00 2.05e+01 1.36e+00 2.50e-01 21 2.21e-03 1.99e-03 5.00e+00 2.15e+01 1.22e+00 2.44e-01 22 2.08e-03 3.26e-03 5.00e+00 2.31e+01 1.47e+00 2.38e-01 23 1.78e-03 2.20e-03 5.00e+00 2.82e+01 1.37e+00 2.33e-01 24 1.90e-03 1.85e-03 5.00e+00 1.86e+01 1.18e+00 2.27e-01 25 2.12e-03 1.93e-03 5.00e+00 2.00e+01 1.12e+00 2.22e-01 26 1.77e-03 2.20e-03 5.00e+00 2.63e+01 1.20e+00 2.17e-01 27 1.62e-03 2.10e-03 5.00e+00 2.55e+01 8.04e-01 2.13e-01 28 1.57e-03 2.09e-03 5.00e+00 2.52e+01 6.65e-01 2.08e-01 29 1.83e-03 1.72e-03 5.00e+00 1.66e+01 1.16e+00 2.04e-01 30 1.68e-03 2.21e-03 5.00e+00 2.46e+01 8.82e-01 2.00e-01 31 1.93e-03 1.76e-03 5.00e+00 1.77e+01 1.11e+00 1.96e-01 32 1.65e-03 1.52e-03 5.00e+00 1.79e+01 8.01e-01 1.92e-01 33 1.85e-03 1.73e-03 5.00e+00 1.54e+01 9.75e-01 1.89e-01 34 1.71e-03 2.25e-03 5.00e+00 2.31e+01 1.11e+00 1.85e-01 35 1.95e-03 1.92e-03 5.00e+00 1.49e+01 8.52e-01 1.82e-01 36 1.66e-03 2.19e-03 5.00e+00 2.22e+01 8.43e-01 1.79e-01 37 1.95e-03 1.98e-03 5.00e+00 1.45e+01 7.60e-01 1.75e-01 38 2.04e-03 1.91e-03 5.00e+00 1.60e+01 9.32e-01 1.72e-01 39 1.62e-03 2.13e-03 5.00e+00 2.13e+01 9.33e-01 1.69e-01 40 2.25e-03 3.39e-03 5.00e+00 1.63e+01 1.45e+00 1.67e-01 41 1.96e-03 3.07e-03 5.00e+00 1.56e+01 1.04e+00 1.64e-01 42 1.81e-03 2.88e-03 5.00e+00 1.54e+01 9.35e-01 1.61e-01 43 1.70e-03 2.72e-03 5.00e+00 1.54e+01 9.57e-01 1.59e-01 44 1.87e-03 1.62e-03 5.00e+00 1.39e+01 8.95e-01 1.56e-01 45 1.98e-03 1.90e-03 5.00e+00 1.27e+01 9.39e-01 1.54e-01 46 1.90e-03 3.08e-03 5.00e+00 1.51e+01 9.98e-01 1.52e-01 47 1.76e-03 2.22e-03 5.00e+00 1.88e+01 1.15e+00 1.49e-01 48 1.71e-03 2.84e-03 5.00e+00 1.47e+01 8.75e-01 1.47e-01 49 1.93e-03 1.89e-03 5.00e+00 1.20e+01 8.04e-01 1.45e-01 ---------------------------------------------------------------- OnlineConvBPDNDictLearn solve time: 55.20s
Display initial and final dictionaries.
D1 = D1.squeeze()
fig = plot.figure(figsize=(14, 7))
plot.subplot(1, 2, 1)
plot.imview(util.tiledict(D0), title='D0', fig=fig)
plot.subplot(1, 2, 2)
plot.imview(util.tiledict(D1), title='D1', fig=fig)
fig.show()
Get iterations statistics from solver object and plot functional value.
its = d.getitstat()
fig = plot.figure(figsize=(7, 7))
plot.plot(np.vstack((its.DeltaD, its.Eta)).T, xlbl='Iterations',
lgnd=('Delta D', 'Eta'), fig=fig)
fig.show()