def main_worker(local_rank, args):
args.local_rank = local_rank
dist.init_process_group(backend='nccl',
rank=local_rank,
world_size=args.world_size)
torch.cuda.set_device(args.local_rank)
# network=nn.DataParallel(Xception(num_classes=cfg.num_classes))
if local_rank == 0:
network = torch.load(args.ckpt_path)
dataloader = create_dataset_pytorch(args.data_path, is_train=False)
with torch.no_grad():
time_start = time.time()
total_samples = 0.0
correct_samples = 0.0
for inputs, labels in dataloader:
inputs = inputs.cuda()
labels = labels.cuda()
outputs = network(inputs)
_, max_index = torch.max(outputs, dim=-1)
total_samples += labels.size(0)
correct_samples += (max_index == labels).sum()
print('Accuracy: {}'.format(correct_samples / total_samples),
flush=True)
time_end = time.time()
time_step = time_end - time_start
print('Finished Testing', flush=True)
print('Finished Testing, Time:%f', time_step, flush=True)
parser.add_argument('--ckpt_path',
type=str,
default="./checkpoint",
help='path where the checkpoint to be saved')
parser.add_argument('--device_id',
type=int,
default=0,
help='device id of GPU. (Default: 0)')
args = parser.parse_args()
device = torch.device('cuda:' + str(args.device_id))
network = torch.load(args.ckpt_path)
network = network.module
network.to(device)
dataloader = create_dataset_pytorch(args.data_path, train=False)
with torch.no_grad():
total_samples = 0.0
correct_samples = 0.0
for inputs, labels in dataloader:
inputs = inputs.to(device)
labels = labels.to(device)
outputs = network(inputs)
_, max_index = torch.max(outputs, dim=-1)
total_samples += labels.size(0)
correct_samples += (max_index == labels).sum()
# print('Accuracy: {}'.format(correct_samples / total_samples), flush=True)
print(correct_samples * 1.0 / total_samples, flush=True)
# print('Finished Testing', flush=True)
help='path where the dataset is saved')
parser.add_argument('--ckpt_path',
type=str,
default="./checkpoint",
help='path where the checkpoint to be saved')
parser.add_argument('--device_id',
type=int,
default=0,
help='device id of GPU. (Default: 0)')
args = parser.parse_args()
device = torch.device('cuda:' + str(args.device_id))
network = torch.load(args.ckpt_path)
network.to(device)
dataloader = create_dataset_pytorch(args.data_path + "/val")
with torch.no_grad():
total_samples = 0.0
correct_samples = 0.0
for inputs, labels in dataloader:
inputs = inputs.to(device)
labels = labels.to(device)
outputs = network(inputs)
_, max_index = torch.max(outputs, dim=-1)
total_samples += labels.size(0)
correct_samples += (max_index == labels).sum()
print('Accuracy: {}'.format(correct_samples / total_samples),
flush=True)
print('Finished Testing', flush=True)
help='path where the dataset is saved')
parser.add_argument('--ckpt_path',
type=str,
default="./checkpoint",
help='path where the checkpoint to be saved')
parser.add_argument('--device_id',
type=int,
default=0,
help='device id of GPU. (Default: 0)')
args = parser.parse_args()
device = torch.device('cuda:' + str(args.device_id))
network = torch.load(args.ckpt_path)
network.to(device)
dataloader = create_dataset_pytorch(args.data_path, is_train=True)
with torch.no_grad():
total_samples = 0.0
correct_samples = 0.0
for inputs, labels in dataloader:
inputs = inputs.to(device)
labels = labels.to(device)
outputs = network(inputs)
_, max_index = torch.max(outputs, dim=-1)
total_samples += labels.size(0)
correct_samples += (max_index == labels).sum()
print('Accuracy: {}'.format(correct_samples / total_samples),
flush=True)
print('Finished Testing', flush=True)
args = parser.parse_args()
network = Xception(num_classes=cfg.num_classes)
network = network.cuda()
network = nn.DataParallel(network)
criterion = nn.CrossEntropyLoss()
# optimizer = optim.RMSprop(network.parameters(),
# lr=cfg.lr_init,
# eps=cfg.rmsprop_epsilon,
# momentum=cfg.rmsprop_momentum,
# alpha=cfg.rmsprop_decay)
optimizer = optim.SGD(network.parameters(),
lr=cfg.lr_init,
momentum=cfg.SGD_momentum)
# prepare data
dataloader = create_dataset_pytorch(args.data_path,
n_workers=cfg.n_workers)
step_per_epoch = len(dataloader)
print("step_per_epoch =", step_per_epoch)
scheduler = optim.lr_scheduler.StepLR(optimizer,
gamma=cfg.lr_decay_rate,
step_size=cfg.lr_decay_epoch *
step_per_epoch)
q_ckpt = Queue(maxsize=cfg.keep_checkpoint_max)
global_step_id = 0
for epoch in range(cfg.epoch_size):
time_epoch = 0.0
torch.cuda.synchronize()
for i, data in enumerate(dataloader):
time_start = time.time()
inputs, labels = data