def main(args):
broden_root = Path(args.root).resolve()
tree = parts.ObjectTree.from_meta_folder(broden_root / 'meta')
encoder_path, decoder_path = utils.upernet_ckpt(broden_root)
model = Head3(tree, encoder_path, decoder_path)
db = upernet_data_pipeline(broden_root)
metrics = partial(parts.BrodenMetricsClbk,
obj_tree=tree,
split_func=lambda o: o)
learn = Learner(
db,
model,
loss_func=Loss(tree),
callback_fns=metrics,
)
learn.split((learn.model.obj_branch, ))
learn.freeze()
utils.fit_and_log(learn,
'object-P.A',
save=args.save,
epochs=20,
lr=1e-2,
pct_start=args.pct_start)
def main(args):
print(args)
arch = pose.nets[args.resnet]
if args.cnn_fix:
instructor = CNNObserver()
bu_c = 0
else:
instructor = SelfObserveInstructor()
bu_c = 16 * 3
root = Path(__file__).resolve().parent.parent / 'LIP'
db = pose.get_data(root, args.size, bs=args.bs)
self_correct = SelfCorrect()
pckh = partial(pose.Pckh,
niter=3,
mean=False,
heatmap_func=self_correct.heatmap_func)
learn = cs.cs_learner(db,
arch,
instructor,
td_c=16,
bu_c=bu_c,
pretrained=False,
embedding=None,
add_td_out=True,
loss_func=self_correct.loss_func,
metrics=self_correct.accuracy,
callback_fns=[pckh, DataTime])
monitor = 'Total_2'
utils.fit_and_log(learn, args, monitor)
def main(args):
broden_root = Path(args.root).resolve()
tree = parts.ObjectTree.from_meta_folder(broden_root / 'meta')
db10 = parts.upernet_data_pipeline(broden_root, csv_file='broden10.csv')
cls10 = obj_sets[args.obj_set]
obj10 = [tree.obj_names.index(name) for name in cls10]
model, instructor = taskmod(broden_root,
tree,
obj_classes=obj10,
full_head=args.full_head)
if args.fill_ones:
model.embeddings.apply(init_ones)
metrics = partial(parts.BinaryBrodenMetrics,
obj_tree=tree,
thresh=0.5,
obj_classes=obj10)
learn = Learner(db10,
model,
loss_func=instructor.loss,
callbacks=[instructor],
callback_fns=metrics)
learn.split((learn.model.embeddings, ))
learn.freeze()
utils.fit_and_log(learn,
'object-P.A',
save=args.save,
epochs=20,
lr=args.lr,
pct_start=args.pct_start,
start_epoch=args.start_epoch)
def main(args):
print(args)
arch = pose.nets[args.resnet]
n = args.niter
instructor = cs.RecurrentInstructor(n)
pckh = partial(pose.Pckh, niter=n)
loss = pose.RecurrentLoss(n)
root = Path(__file__).resolve().parent.parent / 'LIP'
db = pose.get_data(root, args.size, bs=args.bs)
lateral = lateral_types[args.lateral]
learn = cs.cs_learner(db,
arch,
instructor,
td_c=16,
pretrained=False,
embedding=None,
lateral=lateral,
add_td_out=args.add_td_out,
loss_func=loss,
callback_fns=[pckh, DataTime])
monitor = f'Total_{n - 1}' if n > 1 else 'Total'
utils.fit_and_log(learn, args, monitor)
def main(args):
broden_root = Path(args.root).resolve()
tree = parts.ObjectTree.from_meta_folder(broden_root / 'meta')
encoder_path, decoder_path = utils.upernet_ckpt(broden_root)
model = cs_head.CSHead2(tree,
encoder_path,
decoder_path,
hidden=args.hidden)
clbk = cs_head.Head2Clbk(tree)
db = upernet_data_pipeline(broden_root)
metrics = partial(parts.BrodenMetricsClbk,
obj_tree=tree,
split_func=lambda o: (o[0], o[1]))
clbks = [clbk]
if not args.train_bn:
clbks.append(utils.BnFreeze(model.fpn))
learn = Learner(db,
model,
loss_func=cs_head.Head2Loss(tree),
callbacks=clbks,
callback_fns=metrics,
train_bn=args.train_bn)
learn.split((learn.model.embedding, ))
learn.freeze()
utils.fit_and_log(learn,
'object-P.A',
save=args.save,
epochs=20,
lr=1e-2,
pct_start=args.pct_start)
def main(args):
broden_root = Path(args.root).resolve()
tree = parts.ObjectTree.from_meta_folder(broden_root / 'meta')
model, instructor = get_model(broden_root, tree, ops[args.op])
db = upernet_data_pipeline(broden_root)
metrics = partial(parts.BinaryBrodenMetrics, obj_tree=tree, thresh=0.75)
clbks = [instructor]
if not args.train_bn:
clbks.append(utils.BnFreeze(model.fpn))
learn = Learner(db, model, loss_func=instructor.loss, callbacks=clbks, callback_fns=metrics, train_bn=args.train_bn)
learn.split((learn.model.td,))
learn.freeze()
utils.fit_and_log(learn, 'object-P.A', save=args.save, epochs=20, lr=1e-2, pct_start=args.pct_start)
def main(args):
broden_root = Path(args.root).resolve()
tree = parts.ObjectTree.from_meta_folder(broden_root / 'meta')
db = upernet_data_pipeline(broden_root)
model, instructor = taskmod(broden_root, tree)
metrics = partial(parts.BinaryBrodenMetrics, obj_tree=tree, thresh=0.5)
learn = Learner(db,
model,
loss_func=instructor.loss,
callbacks=[instructor],
callback_fns=metrics)
learn.split((learn.model.embeddings, ))
learn.freeze()
utils.fit_and_log(learn,
'object-P.A',
save=args.save,
epochs=20,
lr=1e-3,
pct_start=args.pct_start)
def main(args):
broden_root = Path(args.root).resolve()
db = parts.get_data(broden_root, bs=args.bs)
tree = parts.ObjectTree.from_meta_folder(broden_root / 'meta')
learn = parts.part_learner(db,
models.resnet34,
tree,
pretrained=args.pretrained,
sample_one=args.sample_one)
lr = args.lr
if args.lr0 is not None:
lr = slice(args.lr0, lr)
utils.fit_and_log(learn,
'object-P.A.',
save=args.save,
epochs=args.epochs,
start_epoch=args.start_epoch,
lr=lr,
wd=args.wd,
load=args.load,
no_one_cycle=args.no_one_cycle)