from utils.parse_args import parse_args
from utils.print_easydict import print_easydict
args = parse_args('Deep learning of graph matching evaluation code.')
import importlib
mod = importlib.import_module(cfg.MODULE)
Net = mod.Net
torch.manual_seed(cfg.RANDOM_SEED)
image_dataset = GMDataset(cfg.DATASET_FULL_NAME,
sets='test',
length=cfg.EVAL.SAMPLES,
obj_resize=cfg.PAIR.RESCALE)
dataloader = get_dataloader(image_dataset)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = Net()
model = model.cuda()
model = DataParallel(model, device_ids=range(torch.cuda.device_count()))
if not Path(cfg.OUTPUT_PATH).exists():
Path(cfg.OUTPUT_PATH).mkdir(parents=True)
now_time = datetime.now().strftime('%Y-%m-%d-%H-%M-%S')
with DupStdoutFileManager(
str(Path(cfg.OUTPUT_PATH) /
('eval_log_' + now_time + '.log'))) as _:
print_easydict(cfg)
classes = dataloader.dataset.classes
port=cfg.VISDOM.PORT)
Visdomins.viz.close()
else:
Visdomins = None
torch.manual_seed(cfg.RANDOM_SEED)
pc_dataset = get_datasets(partition='test',
num_points=cfg.DATASET.POINT_NUM,
unseen=cfg.DATASET.UNSEEN,
noise_type=cfg.DATASET.NOISE_TYPE,
rot_mag=cfg.DATASET.ROT_MAG,
trans_mag=cfg.DATASET.TRANS_MAG,
partial_p_keep=cfg.DATASET.PARTIAL_P_KEEP)
dataloader = get_dataloader(pc_dataset)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = Net()
model = model.to(device)
model = DataParallel(model, device_ids=cfg.GPUS)
if not Path(cfg.OUTPUT_PATH).exists():
Path(cfg.OUTPUT_PATH).mkdir(parents=True)
now_time = datetime.now().strftime('%Y-%m-%d-%H-%M-%S')
with DupStdoutFileManager(
str(Path(cfg.OUTPUT_PATH) /
('eval_log_' + now_time + '.log'))) as _:
print_easydict(cfg)
metrics = eval_model(
Net = mod.Net
torch.manual_seed(cfg.RANDOM_SEED)
dataset_len = {'train': cfg.TRAIN.EPOCH_ITERS * cfg.BATCH_SIZE, 'test': cfg.EVAL.SAMPLES}
data_fn = GMDataset
if cfg.MODULE == 'GL.model':
data_fn = GLDataset
image_dataset = {
x: data_fn(cfg.DATASET_FULL_NAME,
sets=x,
length=dataset_len[x],
cls=cfg.TRAIN.CLASS if x == 'train' else None,
obj_resize=cfg.PAIR.RESCALE)
for x in ('train', 'test')}
dataloader = {x: get_dataloader(image_dataset[x], fix_seed=(x == 'test'))
for x in ('train', 'test')}
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = Net()
model = model.cuda()
if cfg.TRAIN.LOSS_FUNC == 'offset':
criterion = RobustLoss(norm=cfg.TRAIN.RLOSS_NORM)
elif cfg.TRAIN.LOSS_FUNC == 'perm':
criterion = CrossEntropyLoss()
else:
raise ValueError('Unknown loss function {}'.format(cfg.TRAIN.LOSS_FUNC))
optimizer = optim.SGD(model.parameters(), lr=cfg.TRAIN.LR, momentum=cfg.TRAIN.MOMENTUM, nesterov=True)
crossval=(x == 'train'),
train_part=(x == 'train'))
for x in ('train', 'test')
}
pc_dataset['val'] = get_datasets(partition='train',
num_points=cfg.DATASET.POINT_NUM,
unseen=cfg.DATASET.UNSEEN,
noise_type=cfg.DATASET.NOISE_TYPE,
rot_mag=cfg.DATASET.ROT_MAG,
trans_mag=cfg.DATASET.TRANS_MAG,
partial_p_keep=cfg.DATASET.PARTIAL_P_KEEP,
crossval=True,
train_part=False)
dataloader = {
x: get_dataloader(pc_dataset[x], shuffle=(x == 'train'))
for x in ('train', 'val', 'test')
}
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = Net()
model = model.cuda()
permLoss = PermLoss()
if cfg.TRAIN.OPTIM == 'SGD':
optimizer = optim.SGD(model.parameters(),
lr=cfg.TRAIN.LR,
momentum=cfg.TRAIN.MOMENTUM,
nesterov=True)
else: