config.py

import os
import torch
import argparse


parser = argparse.ArgumentParser(description='A PyTorch Implementation of Cascade Cost Volume MVSNet')
parser.add_argument('--mode', default='train', help='train or test', choices=['train', 'test', 'profile'])
parser.add_argument('--model', default='mvsnet', help='select model')
parser.add_argument('--device', default='cuda', help='select model')

parser.add_argument('--dataset', default='dtu_yao', help='select dataset')
parser.add_argument('--trainpath', help='train datapath')
parser.add_argument('--testpath', help='test datapath')
parser.add_argument('--trainlist', help='train list')
parser.add_argument('--testlist', help='test list')

parser.add_argument('--epochs', type=int, default=16, help='number of epochs to train')
parser.add_argument('--lr', type=float, default=0.001, help='learning rate')
parser.add_argument('--lrepochs', type=str, default="4,8,10,12:2", help='epoch ids to downscale lr and the downscale rate')
parser.add_argument('--wd', type=float, default=0.0, help='weight decay')

parser.add_argument('--batch_size', type=int, default=1, help='train batch size')
parser.add_argument('--numdepth', type=int, default=192, help='the number of depth values')
parser.add_argument('--interval_scale', type=float, default=1.06, help='the number of depth values')

parser.add_argument('--loadckpt', default=None, help='load a specific checkpoint')
parser.add_argument('--logdir', default='./checkpoints/debug', help='the directory to save checkpoints/logs')
parser.add_argument('--resume', action='store_true', help='continue to train the model')

parser.add_argument('--summary_freq', type=int, default=50, help='print and summary frequency')
parser.add_argument('--save_freq', type=int, default=1, help='save checkpoint frequency')
parser.add_argument('--eval_freq', type=int, default=1, help='eval freq')

parser.add_argument('--seed', type=int, default=1, metavar='S', help='random seed')
parser.add_argument('--pin_m', action='store_true', help='data loader pin memory')
parser.add_argument("--local_rank", type=int, default=0)

parser.add_argument('--share_cr', action='store_true', help='whether share the cost volume regularization')
parser.add_argument('--ndepths', type=str, default="48,32,8", help='ndepths')
parser.add_argument('--depth_inter_r', type=str, default="4,2,1", help='depth_intervals_ratio')
parser.add_argument('--dlossw', type=str, default="0.5,1.0,2.0", help='depth loss weight for different stage')
parser.add_argument('--cr_base_chs', type=str, default="8,8,8", help='cost regularization base channels')
parser.add_argument('--grad_method', type=str, default="detach", choices=["detach", "undetach"], help='grad method')

parser.add_argument('--using_apex', action='store_true', help='using apex, need to install apex')
parser.add_argument('--sync_bn', action='store_true',help='enabling apex sync BN.')
parser.add_argument('--opt-level', type=str, default="O0")
parser.add_argument('--keep-batchnorm-fp32', type=str, default=None)
parser.add_argument('--loss-scale', type=str, default=None)
parser.add_argument('--smooth_lambda', type=float, default=1.0, help="smooth weight")

args = parser.parse_args()
#os.environ["CUDA_VISIBLE_DEVICES"] = args.device
torch.cuda.manual_seed_all(args.seed)
device = torch.device("cuda" if torch.cuda.is_available() else 'cpu')