def main(data_directory, exp_directory, epochs, batch_size):
# Create the deeplabv3 resnet101 model which is pretrained on a subset
# of COCO train2017, on the 20 categories that are present in the Pascal VOC dataset.
model = createDeepLabv3()
model.train()
data_directory = Path(data_directory)
# Create the experiment directory if not present
exp_directory = Path(exp_directory)
if not exp_directory.exists():
exp_directory.mkdir()
# Specify the loss function
criterion = torch.nn.MSELoss(reduction='mean')
# Specify the optimizer with a lower learning rate
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
# Specify the evaluation metrics
metrics = {'f1_score': f1_score, 'auroc': roc_auc_score}
# Create the dataloader
dataloaders = datahandler.get_dataloader_single_folder(
data_directory, batch_size=batch_size)
_ = train_model(model,
criterion,
dataloaders,
optimizer,
bpath=exp_directory,
metrics=metrics,
num_epochs=epochs)
# Save the trained model
torch.save(model, exp_directory / 'weights.pt')
# Specify the loss function
#criterion = torch.nn.MSELoss(reduction='mean')
# Dice/F1 score - https://en.wikipedia.org/wiki/S%C3%B8rensen%E2%80%93Dice_coefficient
criterion = smp.utils.losses.DiceLoss()
# Specify the optimizer with a lower learning rate
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
# Specify the evalutation metrics
# IoU/Jaccard score - https://en.wikipedia.org/wiki/Jaccard_index
metrics = {'f1_score': f1_score, 'auroc': roc_auc_score}
#metrics = [
# smp.utils.metrics.IoU(threshold=0.5),
#]
# Create the dataloader
dataloaders = datahandler.get_dataloader_single_folder(data_dir,
batch_size=batchsize)
trained_model = train_model(model,
criterion,
dataloaders,
optimizer,
bpath=bpath,
metrics=metrics,
num_epochs=epochs)
# Save the trained model
torch.save({'model_state_dict': trained_model.state_dict()},
os.path.join(bpath, 'state_dict.pt'))
torch.save(model, os.path.join(bpath, 'weights.pt'))
(128, 128, 0):5, # Sky
(0, 128, 0):6, # Vegetation
(128, 0, 0):7, # Window
(0, 0, 0):8 # Void
}
mask = cv2.imread(mask_name)
m = np.zeros((mask.shape[0],mask.shape[1]))
for x in range(mask.shape[0]):
for y in range(mask.shape[1]):
for k in mapping:
if np.all(mask[x,y,:]==k):
m[x,y] = mapping[k]
(very slow process)
'''
# Create the dataloader
dataloaders = datahandler.get_dataloader_single_folder(
data_dir, batch_size=batchsize, class_id_list=class_id_list)
trained_model = train_model(model,
criterion,
dataloaders,
optimizer,
bpath=bpath,
metrics=metrics,
num_epochs=epochs)
# Save the trained model
# torch.save({'model_state_dict':trained_model.state_dict()},os.path.join(bpath,'weights'))
torch.save(model, os.path.join(bpath, 'weights.pt'))
# Create the deeplabv3 resnet101 model which is pretrained on a subset of COCO train2017, on the 20 categories that are present in the Pascal VOC dataset.
model = createDeepLabv3(outputConfig)
model.train()
# Create the experiment directory if not present
if not os.path.isdir(bpath):
os.mkdir(bpath)
# Specify the loss function
criterion = torch.nn.MSELoss(reduction='mean')
# Specify the optimizer with a lower learning rate
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
# Specify the evalutation metrics
metrics = {'f1_score': f1_score, 'auroc': roc_auc_score}
# Create the dataloader
dataloaders = datahandler.get_dataloader_single_folder(
data_dir, imageFolder='images', maskFolder='masks', fraction=0.2, batch_size=batchsize)
trained_model = train_model(model, criterion, dataloaders,
optimizer, bpath=bpath, metrics=metrics, num_epochs=epochs)
# Save the trained model
# torch.save({'model_state_dict':trained_model.state_dict()},os.path.join(bpath,'weights'))
torch.save(model, os.path.join(bpath, 'weights.pt'))
model = createDeepLabv3()
model.train()
# Create the experiment directory if not present
if not os.path.isdir(bpath): os.mkdir(bpath)
# Specify the loss function
criterion = torch.nn.MSELoss(reduction='mean')
# Specify the optimizer with a lower learning rate
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
# Specify the evalutation metrics
metrics = {'f1_score': f1_score, 'auroc': roc_auc_score}
# Create the dataloader
dataloaders = datahandler.get_dataloader_single_folder(data_dir,
imageFolder='Image',
maskFolder='Mask',
fraction=0.1,
batch_size=batchsize)
trained_model = train_model(model,
criterion,
dataloaders,
optimizer,
bpath=bpath,
metrics=metrics,
num_epochs=epochs)
# Save the trained model
# torch.save({'model_state_dict':trained_model.state_dict()},os.path.join(bpath,'weights'))
torch.save(model, os.path.join(bpath, 'weights.pt'))
os.mkdir(bpath)
# Specify the loss function
criterion = torch.nn.CrossEntropyLoss()
# Specify the optimizer with a lower learning rate
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
#scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(optimizer, T_0=2)
#scheduler = torch.optim.lr_scheduler.CyclicLR(optimizer, base_lr=1e-5, max_lr=1e-4)
#scheduler = None
# Specify the evalutation metrics
metrics = {'acc': accuracy_score}
# Create the dataloader
dataloaders = datahandler.get_dataloader_single_folder(data_dir,
batch_size=batchsize,
imageFolder='image',
maskFolder='mask')
testloaders = datahandler.get_dataloader_single_folder(val_dir,
batch_size=batchsize,
imageFolder='image',
maskFolder='mask')
trained_model = train_model(model,
criterion,
dataloaders,
testloaders,
optimizer,
bpath=bpath,
metrics=metrics,
num_epochs=epochs)
