help='compressing ration %')
parser.add_argument('--path',
default='wts/unsupervised_ratio10',
help='path to pretrained model')
parser.add_argument('--data', default='BSD68', help='which data to use')
parser.add_argument('--outpath',
default=None,
help='where to save predictions')
opts = parser.parse_args()
ratio = opts.ratio / 100.0
if opts.path.endswith('.npz'):
mfile = opts.path
else:
wts = opts.path
msave = ut.ckpter(wts + '/iter_*.model.npz')
mfile = msave.latest
outpath = opts.outpath
if outpath is not None:
if not os.path.exists(outpath):
os.makedirs(outpath)
VLIST = 'data/%s.txt' % opts.data
PSZ = 33
# Setup Graphs
is_training = tf.placeholder_with_default(False, shape=[])
model = net.Net(is_training)
elif niter >= drop[0] and niter < drop[1]:
return LR / np.sqrt(10.)
else:
return LR / 10.0
return LR
VALFREQ = 2e2
SAVEFREQ = 1e4
MAXITER = drop[-1]
if not os.path.exists(wts):
os.makedirs(wts)
#########################################################################
# Check for saved weights & optimizer states
msave = ut.ckpter(wts + '/iter_*.model.npz')
ssave = ut.ckpter(wts + '/iter_*.state.npz')
ut.logopen(wts+'/train.log')
niter = msave.iter
#########################################################################
# Setup Graphs
is_training = tf.placeholder_with_default(False, shape=[])
model = net.Net(is_training)
# Images loading setup
tset = Dataset(TLIST, KTLIST, BSZ, niter, rand_kernel=False)
vset = Dataset(VLIST, KVLIST, BSZ, 0, isval=True)
batch, swpT, swpV = tvSwap(tset, vset)
imgs, left_kernels, right_kernels, left_ck, right_ck, seeds = batch