if args.watershed is None:
args.watershed = watershed(probs, show_progress=args.show_progress)
ws = args.watershed
thickness = args.thickness
zcrop1 = [0, thickness]
overlaps = [2**i + 1 for i in range(1, 8)]
results_table = zeros([len(args.thresholds), len(range(1, 8))], dtype=bool)
for j, overlap in enumerate(overlaps):
zcrop2 = [thickness - overlap, 2 * thickness - overlap]
# pdb.set_trace()
probs1, ws1 = crop_probs_and_ws(args.xy_crop + zcrop1, probs, ws)
probs2, ws2 = crop_probs_and_ws(args.xy_crop + zcrop2, probs, ws)
g1 = Rag(ws1,
probs1,
show_progress=args.show_progress,
lowmem=args.low_memory)
g2 = Rag(ws2,
probs2,
show_progress=args.show_progress,
lowmem=args.low_memory)
for i, t in enumerate(args.thresholds):
g1.agglomerate(t)
g2.agglomerate(t)
results_table[i, j] = (juicy_center(
g1.segmentation,
2)[..., -overlap / 2].astype(bool) == juicy_center(
g2.segmentation, 2)[..., overlap / 2].astype(bool)).all()
savetxt('debug.txt', results_table, delimiter='\t')
results_table = hstack([array(args.thresholds)[:, newaxis], results_table])
results_table = \
parser.add_argument('--save-node-training-data', metavar='FILE',
help='Save node features and labels to FILE.'
)
parser.add_argument('--node-classifier', metavar='FILE',
help='Train and output a node split classifier.'
)
args = parser.parse_args()
feature_map_function = eval(args.feature_map_function)
if args.load_classifier is not None:
mpf = classifier_probability(eval(args.feature_map_function),
args.load_classifier)
else:
mpf = eval(args.objective_function)
wsg = Rag(args.ws, args.probs, mpf)
features, labels, weights, history, ave_sizes = \
wsg.learn_agglomerate(args.gt, feature_map_function)
print 'shapes: ', features.shape, labels.shape
if args.load_classifier is not None:
try:
f = h5py.File(args.save_training_data)
old_features = array(f['samples'])
old_labels = array(f['labels'])
features = concatenate((features, old_features), 0)
labels = concatenate((labels, old_labels), 0)
except:
pass
print "fitting classifier of size, pos: ", labels.size, (labels==1).sum()
probs = median_filter(probs, 3)
elif args.gaussian_filter is not None:
probs = gaussian_filter(probs, args.gaussian_filter)
if args.watershed is None:
args.watershed = watershed(probs, show_progress=args.show_progress)
ws = args.watershed
thickness = args.thickness
zcrop1 = [0, thickness]
overlaps = [2 ** i + 1 for i in range(1, 8)]
results_table = zeros([len(args.thresholds), len(range(1, 8))], dtype=bool)
for j, overlap in enumerate(overlaps):
zcrop2 = [thickness - overlap, 2 * thickness - overlap]
# pdb.set_trace()
probs1, ws1 = crop_probs_and_ws(args.xy_crop + zcrop1, probs, ws)
probs2, ws2 = crop_probs_and_ws(args.xy_crop + zcrop2, probs, ws)
g1 = Rag(ws1, probs1, show_progress=args.show_progress, lowmem=args.low_memory)
g2 = Rag(ws2, probs2, show_progress=args.show_progress, lowmem=args.low_memory)
for i, t in enumerate(args.thresholds):
g1.agglomerate(t)
g2.agglomerate(t)
results_table[i, j] = (
juicy_center(g1.segmentation, 2)[..., -overlap / 2].astype(bool)
== juicy_center(g2.segmentation, 2)[..., overlap / 2].astype(bool)
).all()
savetxt("debug.txt", results_table, delimiter="\t")
results_table = hstack([array(args.thresholds)[:, newaxis], results_table])
results_table = vstack([array([0] + overlaps), results_table, results_table.all(axis=0)])
savetxt(args.fout, results_table, delimiter="\t", fmt="%i")
parser.add_argument('--save-node-training-data',
metavar='FILE',
help='Save node features and labels to FILE.')
parser.add_argument('--node-classifier',
metavar='FILE',
help='Train and output a node split classifier.')
args = parser.parse_args()
feature_map_function = eval(args.feature_map_function)
if args.load_classifier is not None:
mpf = classifier_probability(eval(args.feature_map_function),
args.load_classifier)
else:
mpf = eval(args.objective_function)
wsg = Rag(args.ws, args.probs, mpf)
features, labels, weights, history, ave_sizes = \
wsg.learn_agglomerate(args.gt, feature_map_function)
print 'shapes: ', features.shape, labels.shape
if args.load_classifier is not None:
try:
f = h5py.File(args.save_training_data)
old_features = array(f['samples'])
old_labels = array(f['labels'])
features = concatenate((features, old_features), 0)
labels = concatenate((labels, old_labels), 0)
except:
pass
print "fitting classifier of size, pos: ", labels.size, (labels == 1).sum()
