def load_model_fpn(_model_weights, is_inference=False):
print("Using weights {}".format(_model_weights))
if _model_weights == "imagenet":
model = FPN(unet_encoder,
encoder_weights="imagenet",
classes=4,
activation=None)
if is_inference:
model.eval()
return model
else:
model = FPN(unet_encoder,
encoder_weights=None,
classes=4,
activation=None)
if is_inference:
model.eval()
if _model_weights is not None:
device = torch.device("cuda")
model.to(device)
state = torch.load(
_model_weights) # , map_location=lambda storage, loc: storage)
model.load_state_dict(state["state_dict"])
# new_state_dict = OrderedDict()
#
# for k, v in state['state_dict'].items():
# if k in model.state_dict():
# new_state_dict[k] = v
# model = model.load_state_dict(new_state_dict)
return model
state = torch.load(
'../input/senetmodels/senext50_30_epochs_high_threshold.pth',
map_location=lambda storage, loc: storage)
model_senet.load_state_dict(state["state_dict"])
model_fpn91lb = FPN(encoder_name="se_resnext50_32x4d",
classes=4,
activation=None,
encoder_weights=None)
model_fpn91lb.to(device)
model_fpn91lb.eval()
#state = torch.load('../input/fpnseresnext/model_se_resnext50_32x4d_fold_0_epoch_7_dice_0.935771107673645.pth', map_location=lambda storage, loc: storage)
state = torch.load(
'../input/fpnse50dice944/model_se_resnext50_32x4d_fold_0_epoch_26_dice_0.94392.pth',
map_location=lambda storage, loc: storage)
model_fpn91lb.load_state_dict(state["state_dict"])
model_fpn91lb_pseudo = FPN(encoder_name="se_resnext50_32x4d",
classes=4,
activation=None,
encoder_weights=None)
model_fpn91lb_pseudo.to(device)
model_fpn91lb_pseudo.eval()
#state = torch.load('../input/fpnseresnext/model_se_resnext50_32x4d_fold_0_epoch_7_dice_0.935771107673645.pth', map_location=lambda storage, loc: storage)
state = torch.load(
'../input/942-finetuned-on-pseudo-to9399/pseudo_fpn_se_resnext50_32x4d_fold_0_epoch_22_dice_0.944/pseudo_fpn_se_resnext50_32x4d_fold_0_epoch_22_dice_0.9446276426315308.pth',
map_location=lambda storage, loc: storage)
model_fpn91lb_pseudo.load_state_dict(state["state_dict"])
ENCODER = 'se_resnext50_32x4d'
ENCODER_WEIGHTS = 'imagenet'
def predict_valid():
inputdir = "/data/Thoracic_OAR/"
transform = valid_aug(image_size=512)
# nii_files = glob.glob(inputdir + "/*/data.nii.gz")
folds = [0, 1, 2, 3, 4]
for fold in folds:
print(fold)
outdir = f"/data/Thoracic_OAR_predict/FPN-seresnext50/"
log_dir = f"/logs/ss_miccai/FPN-se_resnext50_32x4d-fold-{fold}"
# model = VNet(
# encoder_name='se_resnext50_32x4d',
# encoder_weights=None,
# classes=7,
# # activation='sigmoid',
# group_norm=False,
# center='none',
# attention_type='scse',
# reslink=True,
# multi_task=False
# )
model = FPN(encoder_name='se_resnext50_32x4d',
encoder_weights=None,
classes=7)
ckp = os.path.join(log_dir, "checkpoints/best.pth")
checkpoint = torch.load(ckp)
model.load_state_dict(checkpoint['model_state_dict'])
model = nn.DataParallel(model)
model = model.to(device)
df = pd.read_csv(f'./csv/5folds/valid_{fold}.csv')
patient_ids = df.patient_id.unique()
for patient_id in patient_ids:
print(patient_id)
nii_file = f"{inputdir}/{patient_id}/data.nii.gz"
image_slices, ct_image = extract_slice(nii_file)
dataset = TestDataset(image_slices, transform)
dataloader = DataLoader(dataset=dataset,
num_workers=4,
batch_size=8,
drop_last=False)
pred_mask, pred_logits = predict(model, dataloader)
# import pdb
# pdb.set_trace()
pred_mask = np.argmax(pred_mask, axis=1).astype(np.uint8)
pred_mask = SimpleITK.GetImageFromArray(pred_mask)
pred_mask.SetDirection(ct_image.GetDirection())
pred_mask.SetOrigin(ct_image.GetOrigin())
pred_mask.SetSpacing(ct_image.GetSpacing())
# patient_id = nii_file.split("/")[-2]
patient_dir = f"{outdir}/{patient_id}"
os.makedirs(patient_dir, exist_ok=True)
patient_pred = f"{patient_dir}/predict.nii.gz"
SimpleITK.WriteImage(pred_mask, patient_pred)
