def main():
parser = argparse.ArgumentParser()
parser.add_argument('input_path')
parser.add_argument('output_path', nargs='?', default=None)
args = parser.parse_args()
def ensure_output_directory(*p):
root = args.output_path
if root is None:
root = args.input_path
return ensure_path(os.path.join(root, *p),
is_dir=True)
walker = os.walk(args.input_path)
input_dir, shard_subdirs, files = next(walker)
shard_subdirs = sorted(shard_subdirs, key=int)
shard_dirs = map(lambda d: os.path.join(input_dir, d),
shard_subdirs)
state_paths = map(lambda d: os.path.join(d, 'all_Xy.dat'),
shard_dirs)
all_Xy, a = np.load(state_paths[0])
print '<-', a['input_path']
I = plt.imread(a['input_path']).astype(np.float64)[..., :3]
if a['base'] == 'white':
J = np.ones_like(I)
elif a['base'] == 'black':
J = np.zeros_like(I)
else:
head, tail = os.path.split(a['base'])
root, ext = os.path.splitext(tail)
if ext == '.dat':
J = np.load(a.base)
else:
J = plt.imread(a.base).astype(np.float64)[..., :3]
sr = ShardReconstructor(I, a['alpha'], n=a['n'], k=a['k'])
for state_index, state_path in enumerate(state_paths):
print '<-', state_path
all_Xy, _ = np.load(state_path)
J1s = map(lambda t: sr.add_shard_to_reconstruction(J, t[0], t[1]),
all_Xy)
output_directory = ensure_output_directory(shard_subdirs[state_index])
make_visualisations_inplace(map(itemgetter(0), all_Xy),
J1s,
output_directory,
verbose=True)
make_residual_image(I, J1s[-1], output_directory, verbose=True)
J = np.load(os.path.join(shard_dirs[state_index], 'J.dat'))
def main():
parser = argparse.ArgumentParser()
parser.add_argument('input_path')
parser.add_argument('output_dir')
parser.add_argument('base', help='{white, black, BASE}')
parser.add_argument('k', type=float)
parser.add_argument('alpha', type=float)
parser.add_argument('shards', type=int)
parser.add_argument('--n', type=int, default=3)
parser.add_argument('--xtol', type=float, default=1e-4)
parser.add_argument('--epsilon', type=float, default=1e-6)
parser.add_argument('--visualise-progress',
action='store_true',
default=False)
parser.add_argument('--continue-from-previous',
action='store_true',
default=False)
parser.add_argument('--num-restarts-per-shard',
type=int,
default=1)
parser.add_argument('--dont-solve-joint',
dest='solve_joint',
default=True,
action='store_false')
args = parser.parse_args()
ensure_output_path = partial(vis.ensure_path, args.output_dir)
print '<-', args.input_path
I = plt.imread(args.input_path).astype(np.float64)[..., :3]
if args.base == 'white':
J = np.ones_like(I)
elif args.base == 'black':
J = np.zeros_like(I)
else:
head, tail = os.path.split(args.base)
root, ext = os.path.splitext(tail)
if ext == '.dat':
J = np.load(args.base)
else:
J = plt.imread(args.base).astype(np.float64)[..., :3]
sr = ShardReconstructor(I, args.alpha, n=args.n, k=args.k)
if args.continue_from_previous:
for n in xrange(args.shards - 1, -1, -1):
p = os.path.join(args.output_dir, str(n), 'J.dat')
try:
J = np.load(p)
except IOError:
continue
break
n += 1
else:
n = 0
print 'n: [%d, %d)' % (n, args.shards)
t1 = time()
all_iterations_Xy = []
for n in xrange(n, args.shards):
E0 = sr.reconstruction_energy(J)
max_E_diff = 0.0
for i in xrange(args.num_restarts_per_shard):
X, y, all_Xy = sr.candidate_shard(J, args.solve_joint,
verbose=True,
epsilon=args.epsilon,
xtol=args.xtol)
J1 = sr.add_shard_to_reconstruction(J, X, y)
E1 = sr.reconstruction_energy(J1)
E_diff = E0 - E1
if E_diff > max_E_diff:
max_X_y = (X, y, all_Xy)
max_E_diff = E_diff
min_E1 = E1
if max_E_diff == 0.0:
# no shard produced a decrease in energy.
print 'shard %d ignored' % n
continue
X, y, all_Xy = max_X_y
all_iterations_Xy.append(all_Xy)
print 'E:', E0
print 'E(next):', min_E1
J1s = map(lambda t: sr.add_shard_to_reconstruction(J, t[0], t[1]),
all_Xy)
J = J1s[-1]
output_dir = ensure_output_path(n, is_dir=True)
if args.visualise_progress:
vis.make_visualisations_inplace(map(itemgetter(0), all_Xy),
J1s,
output_dir, verbose=True)
vis.make_residual_image(I, J1s[-1],
output_dir, verbose=True)
output_path = ensure_output_path(n, 'all_Xy.dat')
print '->', output_path
dump(output_path, (all_Xy, args.__dict__), raise_on_failure=False)
output_path = ensure_output_path(n, 'all_iterations_Xy.dat')
print '->', output_path
dump(output_path, (all_iterations_Xy, args.__dict__),
raise_on_failure=False)
output_path = ensure_output_path(n, 'J.dat')
print '->', output_path
dump(output_path, J, raise_on_failure=False)
print '%d shards in %.3fs' % (n + 1, time() - t1)