def run(config_file):
config = load_config(config_file)
os.makedirs(config.work_dir, exist_ok=True)
save_config(config, config.work_dir + '/config.yml')
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
all_transforms = {}
all_transforms['train'] = get_transforms(config.transforms.train)
all_transforms['valid'] = get_transforms(config.transforms.test)
dataloaders = {
phase: make_loader(
data_folder=config.data.train_dir,
df_path=config.data.train_df_path,
phase=phase,
batch_size=config.train.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=all_transforms[phase],
num_classes=config.data.num_classes,
pseudo_label_path=config.train.pseudo_label_path,
task='cls'
)
for phase in ['train', 'valid']
}
# create model
model = CustomNet(config.model.encoder, config.data.num_classes)
# train setting
criterion = get_loss(config)
params = [
{'params': model.base_params(), 'lr': config.optimizer.params.encoder_lr},
{'params': model.fresh_params(), 'lr': config.optimizer.params.decoder_lr}
]
optimizer = get_optimizer(params, config)
scheduler = get_scheduler(optimizer, config)
# model runner
runner = SupervisedRunner(model=model)
callbacks = [MultiClassAccuracyCallback(threshold=0.5), F1ScoreCallback()]
if os.path.exists(config.work_dir + '/checkpoints/best.pth'):
callbacks.append(CheckpointCallback(resume=config.work_dir + '/checkpoints/best_full.pth'))
# model training
runner.train(
model=model,
criterion=criterion,
optimizer=optimizer,
scheduler=scheduler,
loaders=dataloaders,
logdir=config.work_dir,
num_epochs=config.train.num_epochs,
callbacks=callbacks,
verbose=True,
fp16=True,
)
def run(config_file, fold=0, device_id=0, ensemble=False):
os.environ['CUDA_VISIBLE_DEVICES'] = str(device_id)
config = load_config(config_file)
if not '_fold' in config.work_dir and not ensemble:
config.work_dir = config.work_dir + '_fold{}'.format(fold)
testloader = make_loader(
data_dir=config.data.test_dir,
df_path=config.data.sample_submission_path,
features=config.data.features,
phase='test',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test),
)
if ensemble:
# load model
models = []
for c in model_config_paths:
for i in range(5):
models.append(load_fold_model(c, i))
model = MultiModels(models, tta=False)
else:
checkpoint_path = config.work_dir + '/checkpoints/best.pth'
model = load_model(config_file, checkpoint_path, fold)
predictions = []
z_pos = config.data.z_pos[0]
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = model(batch_images.to(config.device))
batch_preds[:, 0] = torch.sigmoid(batch_preds[:, 0])
batch_preds[:, z_pos] = depth_transform(batch_preds[:, z_pos])
batch_preds = batch_preds.data.cpu().numpy()
for preds in batch_preds:
coords = extract_coords(
preds,
features=config.data.features,
img_size=(config.data.height, config.data.width),
confidence_threshold=config.test.confidence_threshold,
distance_threshold=config.test.distance_threshold,
)
s = coords2str(coords)
predictions.append(s)
# ---------------------------------------------------------------------------------
# submission
# ------------------------------------------------------------------------------------------------------------
test = pd.read_csv(config.data.sample_submission_path)
test['PredictionString'] = predictions
out_path = config.work_dir + '/submission.csv'
test.to_csv(out_path, index=False)
postprocess(out_path)
def run_cls(config_file_cls):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 1. classification inference
# ------------------------------------------------------------------------------------------------------------
config = load_config(config_file_cls)
validloader = make_loader(data_folder=config.data.train_dir,
df_path=config.data.train_df_path,
phase='valid',
batch_size=config.train.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=get_transforms(
config.transforms.test),
num_classes=config.data.num_classes,
task='cls')
model = CustomNet(config.model.encoder, config.data.num_classes)
model.to(config.device)
model.eval()
checkpoint = load_checkpoint(f"{config.work_dir}/checkpoints/best.pth")
model.load_state_dict(checkpoint['model_state_dict'])
all_predictions = []
all_targets = []
with torch.no_grad():
for i, (batch_images, batch_targets) in enumerate(tqdm(validloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta,
task='cls')
all_targets.append(batch_targets)
all_predictions.append(batch_preds)
all_predictions = np.concatenate(all_predictions)
all_targets = np.concatenate(all_targets)
# evaluation
all_accuracy_scores = []
all_f1_scores = []
thresholds = np.linspace(0.1, 0.9, 9)
for th in thresholds:
accuracy = accuracy_score(all_targets > th, all_predictions > th)
f1 = f1_score(all_targets > th,
all_predictions > th,
average='samples')
all_accuracy_scores.append(accuracy)
all_f1_scores.append(f1)
for th, score in zip(thresholds, all_accuracy_scores):
print('validation accuracy for threshold {} = {}'.format(th, score))
for th, score in zip(thresholds, all_f1_scores):
print('validation f1 score for threshold {} = {}'.format(th, score))
np.save('valid_preds', all_predictions)
def run_cls(config_dir):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 1. classification inference
# ------------------------------------------------------------------------------------------------------------
config_root = Path(config_dir) / 'cls'
config_paths = [config_root / p for p in os.listdir(config_root)]
base_config_paths = [
Path(config_dir) / p for p in os.listdir(config_dir) if 'yml' in p
]
config = load_config(base_config_paths[0])
models = []
for c in config_paths:
models.append(load_model(c))
model = MultiClsModels(models)
testloader = make_loader(
data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test),
num_classes=config.data.num_classes,
)
all_fnames = []
all_predictions = []
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta,
task='cls')
all_fnames.extend(batch_fnames)
all_predictions.append(batch_preds)
all_predictions = np.concatenate(all_predictions)
np.save('all_preds', all_predictions)
df = pd.DataFrame(data=all_predictions, index=all_fnames)
df.to_csv('cls_preds.csv')
df.to_csv(KAGGLE_WORK_DIR + '/cls_preds.csv')
def run_cls(config_file_cls):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 1. classification inference
# ------------------------------------------------------------------------------------------------------------
config = load_config(config_file_cls)
model = CustomNet(config.model.encoder, config.data.num_classes)
testloader = make_loader(
data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test),
num_classes=config.data.num_classes,
)
model.to(config.device)
model.eval()
checkpoint = load_checkpoint(f"{config.work_dir}/checkpoints/best.pth")
model.load_state_dict(checkpoint['model_state_dict'])
all_fnames = []
all_predictions = []
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta,
task='cls')
all_fnames.extend(batch_fnames)
all_predictions.append(batch_preds)
all_predictions = np.concatenate(all_predictions)
np.save('all_preds', all_predictions)
df = pd.DataFrame(data=all_predictions, index=all_fnames)
df.to_csv('cls_preds.csv', index=False)
df.to_csv(f"{config.work_dir}/cls_preds.csv", index=False)
def get_model_and_loader(config_paths):
config = load_config(config_paths[0])
models = []
for c in config_paths:
models.append(load_model(c))
model = MultiSegModels(models)
testloader = make_loader(
data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test))
return model, testloader
def run_seg(config_file_seg):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 2. segmentation inference
# ------------------------------------------------------------------------------------------------------------
config = load_config(config_file_seg)
model = getattr(smp, config.model.arch)(
encoder_name=config.model.encoder,
encoder_weights=config.model.pretrained,
classes=config.data.num_classes,
activation=None,
)
if os.path.exists('cls_preds.csv'):
testloader = make_loader(data_folder=config.data.test_dir,
df_path='cls_preds.csv',
phase='filtered_test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(
config.transforms.test))
else:
testloader = make_loader(data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(
config.transforms.test))
model.to(config.device)
model.eval()
checkpoint = load_checkpoint(f"{config.work_dir}/checkpoints/best.pth")
model.load_state_dict(checkpoint['model_state_dict'])
if os.path.exists(config.work_dir + '/threshold_search.json'):
with open(config.work_dir + '/threshold_search.json') as json_file:
data = json.load(json_file)
df = pd.DataFrame(data)
min_sizes = list(df.T.idxmax().values.astype(int))
print('load best threshold from validation:', min_sizes)
else:
min_sizes = config.test.min_size
print('load default threshold:', min_sizes)
predictions = []
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta)
for fname, preds in zip(batch_fnames, batch_preds):
if config.data.num_classes == 4:
for cls in range(preds.shape[0]):
mask = preds[cls, :, :]
mask, num = post_process(mask,
config.test.best_threshold,
min_sizes[cls])
rle = mask2rle(mask)
name = fname + f"_{cls + 1}"
predictions.append([name, rle])
else: # == 5
for cls in range(1, 5):
mask = preds[cls, :, :]
mask, num = post_process(mask,
config.test.best_threshold,
min_sizes[cls])
rle = mask2rle(mask)
name = fname + f"_{cls}"
predictions.append([name, rle])
# ------------------------------------------------------------------------------------------------------------
# submission
# ------------------------------------------------------------------------------------------------------------
df = pd.DataFrame(predictions,
columns=['ImageId_ClassId', 'EncodedPixels'])
df.to_csv(config.work_dir + "/submission.csv", index=False)
def run_seg(config_file_seg):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 2. segmentation inference
# ------------------------------------------------------------------------------------------------------------
config = load_config(config_file_seg)
if 'COLAB_GPU' in os.environ:
config.work_dir = '/content/drive/My Drive/kaggle_cloud/' + config.work_dir
elif 'KAGGLE_WORKING_DIR' in os.environ:
config.work_dir = '/kaggle/working/' + config.work_dir
if os.path.exists('cls_preds.csv'):
testloader = make_loader(
data_folder=config.data.test_dir,
df_path='cls_preds.csv',
phase='filtered_test',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test))
else:
testloader = make_loader(
data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test))
model = load_model(config_file_seg)
if os.path.exists(config.work_dir + '/threshold_search.json'):
with open(config.work_dir + '/threshold_search.json') as json_file:
data = json.load(json_file)
df = pd.DataFrame(data)
min_sizes = list(df.T.idxmax().values.astype(int))
print('load best threshold from validation:', min_sizes)
else:
min_sizes = config.test.min_size
print('load default threshold:', min_sizes)
predictions = []
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta)
for fname, preds in zip(batch_fnames, batch_preds):
for cls in range(preds.shape[0]):
pred, _ = post_process(
preds[cls, :, :],
config.test.best_threshold,
min_sizes[cls],
height=config.transforms.test.Resize.height,
width=config.transforms.test.Resize.width)
pred = cv2.resize(pred, (SUB_WIDTH, SUB_HEIGHT))
pred = (pred > 0.5).astype(int)
rle = mask2rle(pred)
cls_name = INV_CLASSES[cls]
name = fname + f"_{cls_name}"
predictions.append([name, rle])
# ------------------------------------------------------------------------------------------------------------
# submission
# ------------------------------------------------------------------------------------------------------------
df = pd.DataFrame(predictions, columns=['Image_Label', 'EncodedPixels'])
df.to_csv(config.work_dir + "/submission.csv", index=False)
def run(config_file, fold=0, device_id=0):
os.environ['CUDA_VISIBLE_DEVICES'] = str(device_id)
config = load_config(config_file)
if not '_fold' in config.work_dir:
config.work_dir = config.work_dir + '_fold{}'.format(fold)
validloader = make_loader(
data_dir=config.data.train_dir,
df_path=config.data.train_df_path,
features=config.data.features,
phase='valid',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
idx_fold=fold,
transforms=get_transforms(config.transforms.test),
model_scale=config.data.model_scale,
return_fnames=True,
)
# load model
checkpoint_path = config.work_dir + '/checkpoints/best.pth'
model = load_model(config_file, checkpoint_path)
folds = pd.read_csv('data/folds.csv')
predictions = []
targets = []
image_ids = []
z_pos = config.data.z_pos[0]
with torch.no_grad():
for i, (batch_images, batch_mask_regr,
batch_image_ids) in enumerate(tqdm(validloader)):
batch_preds = model(batch_images.to(config.device))
batch_preds[:, 0] = torch.sigmoid(batch_preds[:, 0])
batch_preds[:, z_pos] = depth_transform(batch_preds[:, z_pos])
batch_preds = batch_preds.data.cpu().numpy()
batch_mask_regr = batch_mask_regr.data.cpu().numpy()
image_ids.extend(batch_image_ids)
for preds, mask_regr, image_id in zip(batch_preds, batch_mask_regr,
batch_image_ids):
coords = extract_coords(
preds,
features=config.data.features,
img_size=(config.data.height, config.data.width),
confidence_threshold=config.test.confidence_threshold,
distance_threshold=config.test.distance_threshold,
)
predictions.append(coords)
s = folds.loc[folds.ImageId == image_id.split('.jpg')[0],
'PredictionString'].values[0]
true_coords = str2coords(
s, names=['id', 'yaw', 'pitch', 'roll', 'x', 'y', 'z'])
targets.append(true_coords)
with open(config.work_dir + '/predictions.pkl', 'wb') as f:
pickle.dump(predictions, f)
with open(config.work_dir + '/targets.pkl', 'wb') as f:
pickle.dump(targets, f)
rows = []
for p, i in zip(predictions, image_ids):
rows.append({'ImageId': i, 'PredictionString': coords2str(p)})
pred_df = pd.DataFrame(rows)
pred_df.to_csv(config.work_dir + '/val_pred.csv', index=False)
all_result, result = calc_map_score(targets, predictions)
result['confidence_threshold'] = config.test.confidence_threshold
result['distance_threshold'] = config.test.distance_threshold
dict_to_json(
all_result, config.work_dir +
'/all_result_th{}.json'.format(config.test.distance_threshold))
dict_to_json(
result, config.work_dir +
'/result_th{}.json'.format(config.test.distance_threshold))
for k in sorted(result.keys()):
print(k, result[k])
def run(config_file):
config = load_config(config_file)
if not os.path.exists(config.work_dir):
os.makedirs(config.work_dir, exist_ok=True)
save_config(config, config.work_dir + '/config.yml')
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
all_transforms = {}
all_transforms['train'] = get_transforms(config.transforms.train)
all_transforms['valid'] = get_transforms(config.transforms.test)
dataloaders = {
phase: make_loader(data_folder=config.data.train_dir,
df_path=config.data.train_df_path,
phase=phase,
batch_size=config.train.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=all_transforms[phase],
num_classes=config.data.num_classes,
pseudo_label_path=config.train.pseudo_label_path,
debug=config.debug)
for phase in ['train', 'valid']
}
# create segmentation model with pre trained encoder
model = getattr(smp, config.model.arch)(
encoder_name=config.model.encoder,
encoder_weights=config.model.pretrained,
classes=config.data.num_classes,
activation=None,
)
# train setting
criterion = get_loss(config)
params = [
{
'params': model.decoder.parameters(),
'lr': config.optimizer.params.decoder_lr
},
{
'params': model.encoder.parameters(),
'lr': config.optimizer.params.encoder_lr
},
]
optimizer = get_optimizer(params, config)
scheduler = get_scheduler(optimizer, config)
# model runner
runner = SupervisedRunner(model=model)
callbacks = [DiceCallback(), IouCallback()]
# to resume from check points if exists
if os.path.exists(config.work_dir + '/checkpoints/best.pth'):
callbacks.append(
CheckpointCallback(resume=config.work_dir +
'/checkpoints/best_full.pth'))
if config.train.mixup:
callbacks.append(MixupCallback())
if config.train.cutmix:
callbacks.append(CutMixCallback())
# model training
runner.train(
model=model,
criterion=criterion,
optimizer=optimizer,
scheduler=scheduler,
loaders=dataloaders,
logdir=config.work_dir,
num_epochs=config.train.num_epochs,
callbacks=callbacks,
verbose=True,
fp16=True,
)
def run_seg(config_file_seg):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 2. segmentation inference
# ------------------------------------------------------------------------------------------------------------
config = load_config(config_file_seg)
validloader = make_loader(
data_folder=config.data.train_dir,
df_path=config.data.train_df_path,
phase='valid',
batch_size=config.train.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=get_transforms(config.transforms.test),
num_classes=config.data.num_classes,
)
# create segmentation model with pre-trained encoder
model = getattr(smp, config.model.arch)(
encoder_name=config.model.encoder,
encoder_weights=config.model.pretrained,
classes=config.data.num_classes,
activation=None,
)
model.to(config.device)
model.eval()
checkpoint = load_checkpoint(f"{config.work_dir}/checkpoints/best.pth")
model.load_state_dict(checkpoint['model_state_dict'])
all_dice = {}
min_sizes = [100, 300, 500, 750, 1000, 1500, 2000, 3000]
for min_size in min_sizes:
all_dice[min_size] = {}
for cls in range(config.data.num_classes):
all_dice[min_size][cls] = []
with torch.no_grad():
for i, (batch_images, batch_masks) in enumerate(tqdm(validloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta)
batch_masks = batch_masks.cpu().numpy()
for masks, preds in zip(batch_masks, batch_preds):
for cls in range(config.data.num_classes):
for min_size in min_sizes:
pred, _ = post_process(preds[cls, :, :],
config.test.best_threshold,
min_size)
mask = masks[cls, :, :]
all_dice[min_size][cls].append(dice_score(pred, mask))
for cls in range(config.data.num_classes):
for min_size in min_sizes:
all_dice[min_size][cls] = sum(all_dice[min_size][cls]) / len(
all_dice[min_size][cls])
dict_to_json(all_dice, config.work_dir + '/threshold_search.json')
if config.data.num_classes == 4:
defect_class = cls + 1
else:
defect_class = cls
print('average dice score for class{} for min_size {}: {}'.format(
defect_class, min_size, all_dice[min_size][cls]))
def run(config_file, device_id, idx_fold):
os.environ['CUDA_VISIBLE_DEVICES'] = str(device_id)
print('info: use gpu No.{}'.format(device_id))
config = load_config(config_file)
# for n-folds loop
if config.data.params.idx_fold == -1:
config.data.params.idx_fold = idx_fold
config.work_dir = config.work_dir + '_fold{}'.format(idx_fold)
elif config.data.params.idx_fold == 0:
original_fold = int(config.work_dir.split('_fold')[1])
if original_fold == idx_fold:
raise Exception(
'if you specify fold 0, you should use train.py or resume from fold 1.'
)
config.data.params.idx_fold = idx_fold
config.work_dir = config.work_dir.split('_fold')[0] + '_fold{}'.format(
idx_fold)
else:
raise Exception('you should use train.py if idx_fold is specified.')
print('info: training for fold {}'.format(idx_fold))
if not os.path.exists(config.work_dir):
os.makedirs(config.work_dir, exist_ok=True)
all_transforms = {}
all_transforms['train'] = get_transforms(config.transforms.train)
all_transforms['valid'] = get_transforms(config.transforms.test)
dataloaders = {
phase: make_loader(
df_path=config.data.train_df_path,
data_dir=config.data.train_dir,
features=config.data.features,
phase=phase,
img_size=(config.data.height, config.data.width),
batch_size=config.train.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=all_transforms[phase],
horizontal_flip=config.train.horizontal_flip,
model_scale=config.data.model_scale,
debug=config.debug,
pseudo_path=config.data.pseudo_path,
)
for phase in ['train', 'valid']
}
# create segmentation model with pre trained encoder
num_features = len(config.data.features)
print('info: num_features =', num_features)
model = CenterNetFPN(
slug=config.model.encoder,
num_classes=num_features,
)
optimizer = get_optimizer(model, config)
scheduler = get_scheduler(optimizer, config)
# model runner
runner = SupervisedRunner(model=model, device=get_device())
# train setting
criterion, callbacks = get_criterion_and_callback(config)
if config.train.early_stop_patience > 0:
callbacks.append(
EarlyStoppingCallback(patience=config.train.early_stop_patience))
if config.train.accumulation_size > 0:
accumulation_steps = config.train.accumulation_size // config.train.batch_size
callbacks.extend(
[OptimizerCallback(accumulation_steps=accumulation_steps)])
# to resume from check points if exists
if os.path.exists(config.work_dir + '/checkpoints/last_full.pth'):
callbacks.append(
CheckpointCallback(resume=config.work_dir +
'/checkpoints/last_full.pth'))
# model training
runner.train(
model=model,
criterion=criterion,
optimizer=optimizer,
scheduler=scheduler,
loaders=dataloaders,
logdir=config.work_dir,
num_epochs=config.train.num_epochs,
main_metric=config.train.main_metric,
minimize_metric=config.train.minimize_metric,
callbacks=callbacks,
verbose=True,
fp16=config.train.fp16,
)
image_path_list=valid_data[0],
label_list=valid_data[1],
transform=valid_transform(x_size=x_size,
y_size=y_size,
mean=mean,
std=std))
test_dataset = datasets.NSvalidationimageData(image_path_list=test_data[0],
label_list=test_data[1],
transform=valid_transform(
x_size=x_size,
y_size=y_size,
mean=mean,
std=std))
train_loader = make_loader(dataset=train_dataset, batch_size=train_batch)
valid_loader = make_loader(dataset=valid_dataset, batch_size=valid_batch)
test_loader = make_loader(dataset=test_dataset, batch_size=valid_batch)
#model
net = torch_models.Network(block=block,
input_shape=(3, y_size, x_size),
fc_shapes=fc_shapes,
n_classes=n_classes)
net.train()
update_params = [] if not fine_tuning else net.parameters()
def validation(config_file_seg):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
config = load_config(config_file_seg)
if 'COLAB_GPU' in os.environ:
config.work_dir = '/content/drive/My Drive/kaggle_cloud/' + config.work_dir
elif 'KAGGLE_WORKING_DIR' in os.environ:
config.work_dir = '/kaggle/working/' + config.work_dir
validloader = make_loader(
data_folder=config.data.train_dir,
df_path=config.data.train_df_path,
phase='valid',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=get_transforms(config.transforms.test),
num_classes=config.data.num_classes,
)
model = load_model(config_file_seg)
min_sizes = np.arange(0, 20000, 5000)
label_thresholds = [0.6, 0.7, 0.8]
mask_thresholds = [0.2, 0.3, 0.4]
all_dice = np.zeros(
(4, len(label_thresholds), len(mask_thresholds), len(min_sizes)))
count = 0
with torch.no_grad():
for i, (batch_images, batch_masks) in enumerate(tqdm(validloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta)
batch_labels = torch.nn.functional.adaptive_max_pool2d(
torch.sigmoid(torch.Tensor(batch_preds)),
1).view(batch_preds.shape[0], -1)
batch_masks = batch_masks.cpu().numpy()
batch_labels = batch_labels.cpu().numpy()
batch_masks = resize_batch_images(batch_masks, SUB_HEIGHT,
SUB_WIDTH)
batch_preds = resize_batch_images(batch_preds, SUB_HEIGHT,
SUB_WIDTH)
for labels, masks, preds in zip(batch_labels, batch_masks,
batch_preds):
for cls in range(config.data.num_classes):
for i, label_th in enumerate(label_thresholds):
for j, mask_th in enumerate(mask_thresholds):
for k, min_size in enumerate(min_sizes):
if labels[cls] <= label_th:
pred = np.zeros(preds[cls, :, :].shape)
else:
pred, _ = post_process(preds[cls, :, :],
mask_th,
min_size,
height=SUB_HEIGHT,
width=SUB_WIDTH)
mask = masks[cls, :, :]
dice = dice_score(pred, mask)
all_dice[cls, i, j, k] += dice
count += 1
all_dice = all_dice / (count)
np.save('all_dice', all_dice)
parameters = {}
parameters['label_thresholds'] = []
parameters['mask_thresholds'] = []
parameters['min_sizes'] = []
parameters['dice'] = []
cv_score = 0
for cls in range(4):
i, j, k = np.where((all_dice[cls] == all_dice[cls].max()))
parameters['label_thresholds'].append(float(label_thresholds[i[0]]))
parameters['mask_thresholds'].append(float(mask_thresholds[j[0]]))
parameters['min_sizes'].append(int(min_sizes[k[0]]))
parameters['dice'].append(float(all_dice[cls].max()))
cv_score += all_dice[cls].max() / 4
print('cv_score:', cv_score)
dict_to_json(parameters, config.work_dir + '/parameters.json')
print(pd.DataFrame(parameters))
def run_seg(config_dir):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 2. segmentation inference
# ------------------------------------------------------------------------------------------------------------
config_root = Path(config_dir) / 'seg'
config_paths = [config_root / p for p in os.listdir(config_root)]
base_config_paths = [
Path(config_dir) / p for p in os.listdir(config_dir) if 'yml' in p
]
config = load_config(base_config_paths[0])
models = []
for c in config_paths:
models.append(load_model(c))
model = MultiSegModels(models)
if os.path.exists('cls_preds.csv'):
testloader = make_loader(data_folder=config.data.test_dir,
df_path='cls_preds.csv',
phase='filtered_test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(
config.transforms.test))
else:
testloader = make_loader(data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(
config.transforms.test))
if os.path.exists(config.work_dir + '/threshold_search.json'):
with open(config.work_dir + '/threshold_search.json') as json_file:
data = json.load(json_file)
df = pd.DataFrame(data)
min_sizes = list(df.T.idxmax().values.astype(int))
print('load best threshold from validation:', min_sizes)
else:
min_sizes = config.test.min_size
print('load default threshold:', min_sizes)
predictions = []
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta)
for fname, preds in zip(batch_fnames, batch_preds):
for cls in range(preds.shape[0]):
mask = preds[cls, :, :]
mask, num = post_process(mask, config.test.best_threshold,
min_sizes[cls])
rle = mask2rle(mask)
name = fname + f"_{cls + 1}"
predictions.append([name, rle])
# ------------------------------------------------------------------------------------------------------------
# submission
# ------------------------------------------------------------------------------------------------------------
sub_df = pd.DataFrame(predictions,
columns=['ImageId_ClassId', 'EncodedPixels'])
sample_submission = pd.read_csv(config.data.sample_submission_path)
df_merged = pd.merge(sample_submission,
sub_df,
on='ImageId_ClassId',
how='left')
df_merged.fillna('', inplace=True)
df_merged['EncodedPixels'] = df_merged['EncodedPixels_y']
df_merged = df_merged[['ImageId_ClassId', 'EncodedPixels']]
df_merged.to_csv("submission.csv", index=False)
df_merged.to_csv(KAGGLE_WORK_DIR + "/submission.csv", index=False)
def run(config_file):
config = load_config(config_file)
if 'COLAB_GPU' in os.environ:
config.work_dir = '/content/drive/My Drive/kaggle_cloud/' + config.work_dir
elif 'KAGGLE_WORKING_DIR' in os.environ:
config.work_dir = '/kaggle/working/' + config.work_dir
print('working directory:', config.work_dir)
if not os.path.exists(config.work_dir):
os.makedirs(config.work_dir, exist_ok=True)
save_config(config, config.work_dir + '/config.yml')
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
all_transforms = {}
all_transforms['train'] = get_transforms(config.transforms.train)
all_transforms['valid'] = get_transforms(config.transforms.test)
dataloaders = {
phase: make_loader(
data_folder=config.data.train_dir,
df_path=config.data.train_df_path,
phase=phase,
img_size=(config.data.height, config.data.width),
batch_size=config.train.batch_size,
num_workers=config.num_workers,
idx_fold=config.data.params.idx_fold,
transforms=all_transforms[phase],
num_classes=config.data.num_classes,
pseudo_label_path=config.train.pseudo_label_path,
debug=config.debug
)
for phase in ['train', 'valid']
}
# create segmentation model with pre trained encoder
model = getattr(smp, config.model.arch)(
encoder_name=config.model.encoder,
encoder_weights=config.model.pretrained,
classes=config.data.num_classes,
activation=None,
)
# train setting
criterion = get_loss(config)
params = [
{'params': model.decoder.parameters(), 'lr': config.optimizer.params.decoder_lr},
{'params': model.encoder.parameters(), 'lr': config.optimizer.params.encoder_lr},
]
optimizer = get_optimizer(params, config)
scheduler = get_scheduler(optimizer, config)
# model runner
runner = SupervisedRunner(model=model, device=get_device())
callbacks = [DiceCallback(), IouCallback()]
if config.train.early_stop_patience > 0:
callbacks.append(EarlyStoppingCallback(
patience=config.train.early_stop_patience))
if config.train.accumulation_size > 0:
accumulation_steps = config.train.accumulation_size // config.train.batch_size
callbacks.extend(
[CriterionCallback(),
OptimizerCallback(accumulation_steps=accumulation_steps)]
)
# to resume from check points if exists
if os.path.exists(config.work_dir + '/checkpoints/best.pth'):
callbacks.append(CheckpointCallback(
resume=config.work_dir + '/checkpoints/last_full.pth'))
if config.train.mixup:
callbacks.append(MixupCallback())
if config.train.cutmix:
callbacks.append(CutMixCallback())
# model training
runner.train(
model=model,
criterion=criterion,
optimizer=optimizer,
scheduler=scheduler,
loaders=dataloaders,
logdir=config.work_dir,
num_epochs=config.train.num_epochs,
main_metric=config.train.main_metric,
minimize_metric=config.train.minimize_metric,
callbacks=callbacks,
verbose=True,
fp16=True,
)
