if args.config_file != "":
cfg.merge_from_file(args.config_file)
cfg.merge_from_list(args.opts)
#cfg.freeze() #skip this, cfg can be modified
gl._init()
gl.set_value('cfg', cfg)
set_seed(cfg.MISC.SEED)
#writer = SummaryWriter()
#gl.set_value('writer', writer)
output_dir = cfg.MISC.OUT_DIR
if output_dir and not os.path.exists(output_dir):
os.makedirs(output_dir)
logger = call_logger(osp.join(output_dir, cfg.MISC.LOGFILE))
gl.set_value('logger', logger)
logger.info("Running with config:\n{}".format(cfg))
# make dataloader
train_loader_all, valid_loader_all = make_data_loader(cfg)
# prepare model
model = build_model(cfg)
torch.save(model.state_dict(),
osp.join(output_dir, f"{cfg.MODEL.NAME}-init.pth"))
# make loss
criterion = make_loss(cfg)
parser.add_argument('--tta', default=0, type=int,help='if tta')
args = parser.parse_args()
if not osp.exists(args.datasets):
raise ValueError(f"Dataset Folder not exist")
#logger file
os.makedirs(args.out_dir,exist_ok = True)
logger = call_logger(osp.join(args.out_dir, args.log_file))
set_seed(args.seed)
#%%model
if 'metatwoview' in args.net:
model = ISICModel_meta(n_class = args.n_class,arch = args.net )
elif 'metasingleview' in args.net:
model = ISICModel_singleview_meta(n_class = args.n_class,arch = args.net )
parser.add_argument(
'--loss_type', default='ce', type=str, help='loss_func'
) # ce/ ce_smooth/bce/bce_smmoth/focal_loss_bce/focal_loss
parser.add_argument('--fn_csv',
default='out.csv',
type=str,
help='out csv file')
args = parser.parse_args()
if args.is_valid and not osp.exists(args.datasets):
raise ValueError(f"Dataset Folder not exist")
#logger file
logger = call_logger(args.log_file)
#%%model
model = models.ISICModel(n_class=args.n_class, arch=args.net)
model.init()
model.load_state_dict(torch.load(args.model_file))
configs = model.configs
#%% augmentation dataset and dataloader
if args.is_tta:
test_transform = TrainAugmentation_albu(size=configs.image_size,
mean=configs.image_mean,
std=configs.image_std)
else:
# return str(Path(out_path)/(Path(dataset.file_path).stem + suffix + '.h5'))
# #return f'{os.path.splitext(dataset.file_path)[0]}{suffix}.h5'
if __name__ == '__main__':
# Load configuration
config = load_config()
# Load model state
model_path = config['model_path']
model_fd = Path(model_path).parent
logger = call_logger(log_file = str(model_fd/'test_log.txt'),log_name = 'UNetPredict')
# Create the model
model = get_model(config)
if 'output_path' in config.keys():
out_path = config['output_path']
else:
out_path = str(model_fd/'h5_pred')
os.makedirs(out_path,exist_ok = True)
logger.info(f'Loading model from {model_path}...')
utils.load_checkpoint(model_path, model)
logger.info(f"Sending the model to '{config['device']}'")
model = model.to(config['device'])
from transforms.data_preprocessing import TrainAugmentation_bbox_albu, TestAugmentation_bbox_albu
from dataset.custom_dataset import CustomDataset_bbox, CustomDataset_bbox_centernet
from torch.utils.data import DataLoader
import torch.nn.functional as F
from torch.optim import SGD
from optim import build_optimizer
from torch.optim.lr_scheduler import MultiStepLR, CyclicLR
from torch.nn.utils import clip_grad_norm_
from tools.loggers import call_logger
from utils.utils import AvgerageMeter
log_file = 'log_locate1.txt'
logger = call_logger(log_file)
def train(loader,
net,
criterion,
optimizer,
device,
epoch,
scheduler=None,
net_type='resnet50_c3_locate'):
net.train(True)
optimizer.zero_grad()
if net_type == 'resnet34_c3_locate' or net_type == 'resnet50_c3_locate':
from pathlib import Path
import os
import cv2
import pydicom
import h5py
from scipy.ndimage import binary_dilation
from pathlib import Path
import os.path as osp
import pandas as pd
from skimage.morphology import skeletonize_3d
from scipy import signal, ndimage
from skimage import measure
from tools.loggers import call_logger
logger = call_logger('CTraw/gen.log', 'GenData')
fd_all = 'CTraw/19_ct'
mask_h5 = 'CTraw/h5/h5_19'
h5_fd = './data/h5'
debugim_fd = './data/debug_img'
name_chs = ['V', 'A', 'PV']
n_class = 2
w_class = [1.0, 0.2] #second term not used
w_class_in_roi = [2.0, 2.0]
df_pid = pd.read_csv('./CTraw/19_ct.csv',
dtype={
'patient': str,
'value': int,
import SimpleITK as sitk
import numpy as np
from pathlib import Path
import os
import cv2
import pydicom
import h5py
from scipy.ndimage import binary_dilation
from tools.loggers import call_logger
fn1 = 'resources/CT/01/data/A_1.mha'
fn2 = 'resources/CT/01/data/PV_1.mha'
dicom_fd = './resources/CT/01/data/'
name_chs = ['V', 'A', 'PV']
out_h5py = './data/h5_rs_one_case/ct01m_c1.h5'
logger = call_logger(log_file=None, 'GenData')
n_class = 2
w_class = [1.0, 0.2]
w_class_in_roi = [2.0, 2.0]
itkimage1 = sitk.ReadImage(fn1)
labels1 = (sitk.GetArrayFromImage(itkimage1)[::-1, :, :] == 1).astype(
'int') #dongmai
#labels0 = (1-labels1).astype('int')
labels1_dilate = binary_dilation(labels1, structure=np.ones(
(5, 5, 5))).astype(labels1.dtype)
# Create main logger
#logger = get_logger('UNet3DTrainer')
#Load and log experiment configuration
config = load_config()
chkpt_sav = config['trainer']['checkpoint_dir']
path_log = Path('../checkpoints') / chkpt_sav
config['trainer']['checkpoint_dir'] = str(path_log)
os.makedirs(path_log, exist_ok=True)
#set seed
manual_seed = config.get('manual_seed', 0)
set_seed(manual_seed)
if config['findlr']:
logger = call_logger(log_file=None, log_name='UNet3DTrainer')
else:
logger = call_logger(log_file=str(
(Path('../checkpoints') / chkpt_sav) / 'log.txt'),
log_name='UNet3DTrainer')
logger.info(config)
manual_seed = config.get('manual_seed', None)
if manual_seed is not None:
logger.info(f'Seed the RNG for all devices with {manual_seed}')
torch.manual_seed(manual_seed)
# see https://pytorch.org/docs/stable/notes/randomness.html
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
