def test(config):
log_dir = os.path.join(config.log_dir, config.name + '_stage_2')
val_path = os.path.join(config.data, "*/test")
val_dataset = MultiviewImgDataset(val_path,
scale_aug=False,
rot_aug=False,
num_views=config.num_views)
val_loader = torch.utils.data.DataLoader(
val_dataset,
batch_size=config.stage2_batch_size,
shuffle=False,
num_workers=0)
pretraining = not config.no_pretraining
cnet = SVCNN(config.name,
nclasses=config.num_classes,
cnn_name=config.cnn_name,
pretraining=pretraining)
cnet_2 = MVCNN(config.name,
cnet,
nclasses=config.num_classes,
cnn_name=config.cnn_name,
num_views=config.num_views)
cnet_2.load(
os.path.join(log_dir, config.snapshot_prefix + str(config.weights)))
optimizer = optim.Adam(cnet_2.parameters(),
lr=config.learning_rate,
weight_decay=config.weight_decay,
betas=(0.9, 0.999))
trainer = ModelNetTrainer(cnet_2,
None,
val_loader,
optimizer,
nn.CrossEntropyLoss(),
config,
log_dir,
num_views=config.num_views)
labels, predictions = trainer.update_validation_accuracy(config.weights,
test=True)
import Evaluation_tools as et
eval_file = os.path.join(config.log_dir, '{}.txt'.format(config.name))
et.write_eval_file(config.data, eval_file, predictions, labels,
config.name)
et.make_matrix(config.data, eval_file, config.log_dir)
train_dataset = SingleImgDataset(args.train_path, scale_aug=False, rot_aug=False, num_models=n_models_train, num_views=args.num_views)
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=64, shuffle=True, num_workers=0)
val_dataset = SingleImgDataset(args.val_path, scale_aug=False, rot_aug=False, test_mode=True)
val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=64, shuffle=False, num_workers=0)
print('num_train_files: '+str(len(train_dataset.filepaths)))
print('num_val_files: '+str(len(val_dataset.filepaths)))
trainer = ModelNetTrainer(cnet, train_loader, val_loader, optimizer, nn.CrossEntropyLoss(), 'svcnn', log_dir, num_views=1)
trainer.train(30)
# STAGE 2
log_dir = args.name+'_stage_2'
create_folder(log_dir)
cnet_2 = MVCNN(args.name, cnet, nclasses=40, cnn_name=args.cnn_name, num_views=args.num_views)
del cnet
optimizer = optim.Adam(cnet_2.parameters(), lr=args.lr, weight_decay=args.weight_decay, betas=(0.9, 0.999))
train_dataset = MultiviewImgDataset(args.train_path, scale_aug=False, rot_aug=False, num_models=n_models_train, num_views=args.num_views)
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batchSize, shuffle=False, num_workers=0)# shuffle needs to be false! it's done within the trainer
val_dataset = MultiviewImgDataset(args.val_path, scale_aug=False, rot_aug=False, num_views=args.num_views,test_mode=True)
val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=args.batchSize, shuffle=False, num_workers=0)
print('num_train_files: '+str(len(train_dataset.filepaths)))
print('num_val_files: '+str(len(val_dataset.filepaths)))
trainer = ModelNetTrainer(cnet_2, train_loader, val_loader, optimizer, nn.CrossEntropyLoss(), 'mvcnn', log_dir, num_views=args.num_views)
trainer.train(30)
def train(config):
log(config.log_file, 'Starting...')
pretraining = not config.no_pretraining
log_dir = config.name
create_folder(config.name)
log(config.log_file, '--------------stage 1--------------')
# STAGE 1
log_dir = os.path.join(config.log_dir, config.name + '_stage_1')
create_folder(log_dir)
cnet = SVCNN(config, pretraining=pretraining)
optimizer = optim.Adam(cnet.parameters(),
lr=config.learning_rate,
weight_decay=config.weight_decay)
train_path = os.path.join(config.data, "*/train")
train_dataset = SingleImgDataset(train_path,
config,
scale_aug=False,
rot_aug=False)
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=config.stage1_batch_size,
shuffle=True,
num_workers=0)
val_path = os.path.join(config.data, "*/test")
val_dataset = SingleImgDataset(val_path,
config,
scale_aug=False,
rot_aug=False,
test_mode=True)
val_loader = torch.utils.data.DataLoader(
val_dataset,
batch_size=config.stage1_batch_size,
shuffle=False,
num_workers=0)
log(config.log_file,
'num_train_files: ' + str(len(train_dataset.filepaths)))
log(config.log_file, 'num_val_files: ' + str(len(val_dataset.filepaths)))
trainer = ModelNetTrainer(cnet,
train_loader,
val_loader,
optimizer,
nn.CrossEntropyLoss(),
config,
log_dir,
num_views=1)
trainer.train(config, config.stage1_batch_size)
#cnet.load(os.path.join(log_dir, config.snapshot_prefix + str(30)))
# STAGE 2
log(config.log_file, '--------------stage 2--------------')
log_dir = os.path.join(config.log_dir, config.name + '_stage_2')
create_folder(log_dir)
cnet_2 = MVCNN(cnet, config)
del cnet
optimizer = optim.Adam(cnet_2.parameters(),
lr=config.learning_rate,
weight_decay=config.weight_decay,
betas=(0.9, 0.999))
train_dataset = MultiviewImgDataset(train_path,
config,
scale_aug=False,
rot_aug=False)
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=config.stage2_batch_size,
shuffle=False,
num_workers=0
) # shuffle needs to be false! it's done within the trainer
val_dataset = MultiviewImgDataset(val_path,
config,
scale_aug=False,
rot_aug=False)
val_loader = torch.utils.data.DataLoader(
val_dataset,
batch_size=config.stage2_batch_size,
shuffle=False,
num_workers=0)
log(config.log_file,
'num_train_files: ' + str(len(train_dataset.filepaths)))
log(config.log_file, 'num_val_files: ' + str(len(val_dataset.filepaths)))
trainer = ModelNetTrainer(cnet_2,
train_loader,
val_loader,
optimizer,
nn.CrossEntropyLoss(),
config,
log_dir,
num_views=config.num_views)
trainer.train(config, config.stage2_batch_size)
elif args.svcnn_training_mode == 'load':
cnet.cuda()
log_dir = args.prefix + 'ckpt/svcnn/model-00025.pth'
# log_dir = ckpt_dir + '/20/mvcnn_stage_1/mvcnn/model-00025.pth'
model = torch.load(log_dir)
cnet.load_state_dict(model)
print('SVCNN trained model loaded!')
# STAGE 2
log_dir = run_dir + '/' + args.name + '_stage_2'
create_folder(log_dir)
cnet_2 = MVCNN(args.name, cnet, nclasses=40, cnn_name=args.cnn_name, num_views=args.num_views, constraint=args.constraint, w_m=args.w_m, T=args.T, preType=args.preType)
del cnet
new_params = list(cnet_2.main_net.parameters()) + list(cnet_2.main_net.parameters()) if args.preType != None else cnet_2.main_net.parameters()
params = new_params if args.freeze else cnet_2.parameters()
optimizer = optim.Adam(params, lr=args.lr/8*args.batchSize, weight_decay=args.weight_decay, betas=(0.9, 0.999))
train_dataset = MultiviewImgDataset(args.train_path, scale_aug=False, rot_aug=False, num_models=n_models_train, num_views=args.num_views)
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batchSize, shuffle=False, num_workers=0) # shuffle needs to be false! it's done within the trainer
val_dataset = MultiviewImgDataset(args.val_path, scale_aug=False, rot_aug=False, num_views=args.num_views)
val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=args.batchSize, shuffle=False, num_workers=0)
print('num_train_files: '+str(len(train_dataset.filepaths)))
print('num_val_files: '+str(len(val_dataset.filepaths)))
trainer = ModelNetTrainer(cnet_2, train_loader, val_loader, optimizer, nn.CrossEntropyLoss(), 'mvcnn', log_dir, num_views=args.num_views)
trainer.train(30)
# STAGE 2
print('###################Stage 2####################')
log_dir = args.name + '_stage_2'
create_folder(log_dir)
# cnet_2与cnet采用相同的网络
cnet_2 = MVCNN(args.name,
cnet,
pool_mode='no',
nclasses=40,
cnn_name=args.cnn_name,
num_views=args.num_views)
del cnet
cnet_2 = torch.nn.DataParallel(cnet_2, device_ids=[0]) # 0,1
optimizer = optim.Adam(cnet_2.parameters(),
lr=args.lr,
weight_decay=args.weight_decay,
betas=(0.9, 0.999))
# bitch_size: 原文为8,内存不足改为6
train_dataset = MultiviewImgDataset(args.train_path,
scale_aug=False,
rot_aug=False,
num_models=n_models_train,
num_views=args.num_views)
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=args.batchSize,
shuffle=False,
'svcnn',
log_dir,
num_views=1)
trainer.train(20)
# STAGE 2
log_dir = 'smtcloss_wonor_' + args.name + '_stage_2_' + localtime
create_folder(log_dir)
cnet_2 = MVCNN(args.name,
cnet,
nclasses=40,
cnn_name=args.cnn_name,
num_views=args.num_views)
del cnet
optimizer = optim.SGD(cnet_2.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
train_dataset = MultiviewImgDataset(args.train_path,
scale_aug=False,
rot_aug=False,
num_models=n_models_train,
num_views=args.num_views)
train_loader = torch.utils.data.DataLoader(
train_dataset, batch_size=args.batchSize, shuffle=False, num_workers=0
) # shuffle needs to be false! it's done within the trainer
val_dataset = MultiviewImgDataset(args.val_path,
scale_aug=False,
rot_aug=False,
