def main():
# Train args
train_opt = TrainOptions()
train_args = train_opt.parse()
# Validation args
val_opt = ValOptions()
val_args = val_opt.parse()
val_args.batchsize = 1
val_args.thread = 0
print('Using PyTorch version: ', torch.__version__, torch.version.cuda)
ngpus_per_node = torch.cuda.device_count()
train_args.world_size = ngpus_per_node * train_args.nnodes
val_args.world_size = ngpus_per_node * train_args.nnodes
train_args.distributed = ngpus_per_node > 1
# Randomize args.dist_url to avoid conflicts on same machine
train_args.dist_url = train_args.dist_url + str(os.getpid() % 100).zfill(2)
if train_args.distributed:
mp.spawn(main_worker,
nprocs=ngpus_per_node,
args=(ngpus_per_node, train_args, val_args))
else:
main_worker(0, ngpus_per_node, train_args, val_args)
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
ROOT_DIR = os.path.dirname(BASE_DIR)
sys.path.append(BASE_DIR)
sys.path.append(ROOT_DIR)
from tools.parse_arg_val import ValOptions
from data.load_dataset import CustomerDataLoader
from lib.models.depth_normal_model import DepthNormal
from lib.utils.net_tools import load_ckpt
from lib.utils.evaluate_depth_error import evaluate_err
from lib.utils.net_tools import save_images
from lib.utils.logging import setup_logging, SmoothedValue
logger = setup_logging(__name__)
if __name__ == '__main__':
test_args = ValOptions().parse()
test_args.thread = 1 # test code only supports thread = 1
test_args.batchsize = 1 # test code only supports batchSize = 1
data_loader = CustomerDataLoader(test_args)
test_datasize = len(data_loader)
logger.info('{:>15}: {:<30}'.format('test_data_size', test_datasize))
# load model
model = DepthNormal()
# evaluate mode
model.eval()
# load checkpoint
if test_args.load_ckpt:
load_ckpt(test_args, model)
model.cuda()
smoothed_criteria = validate_rel_depth_err(pred_depth_metric, data['B_raw'], smoothed_criteria, scale=1.0)
return {'abs_rel': smoothed_criteria['err_absRel'].GetGlobalAverageValue(),
'whdr': smoothed_criteria['err_whdr'].GetGlobalAverageValue()}
if __name__=='__main__':
# Train args
train_opt = TrainOptions()
train_args = train_opt.parse()
merge_cfg_from_file(train_args)
gpu_num = torch.cuda.device_count()
cfg.TRAIN.GPU_NUM = gpu_num
# Validation args
val_opt = ValOptions()
val_args = val_opt.parse()
val_args.batchsize = 1
val_args.thread = 0
# Logger
log_output_dir = cfg.TRAIN.LOG_DIR
if log_output_dir:
try:
os.makedirs(log_output_dir)
except OSError as e:
if e.errno != errno.EEXIST:
raise
logger = setup_logger("lib", log_output_dir, cfg.TRAIN.RUN_NAME + '.txt')
# tensorboard logger
if __name__=='__main__':
# Train args
train_opt = TrainOptions()
train_args = train_opt.parse()
# train_opt.print_options(train_args)
# train_args.load_ckpt="/home/colin/papercode/VNL_Monocular_Depth_Prediction-master/tools/outputs/Jan05-08-40-33_colin-Alienware-Aurora-R7/ckpt/epoch0_step10.pth"
# train_args.resume=True
merge_cfg_from_file(train_args)
# Validation args
val_opt = ValOptions()
val_args = val_opt.parse()
val_args.batchsize = 1
val_args.thread = 0
# val_opt.print_options(val_args)
train_dataloader = CustomerDataLoader(train_args)
train_datasize = len(train_dataloader)
val_dataloader = CustomerDataLoader(val_args)
val_datasize = len(val_dataloader)
gpu_num = torch.cuda.device_count()
# Print configs
print_configs(cfg)