Finetuning¶

class eztorch.models.finetuning.FinetuningModel(trunk, classifier, optimizer, pretrained_trunk_path, trunk_pattern='^(trunk\\\\.)', two_groups=False, freeze_trunk=False, train_transform=None, val_transform=None, test_transform=None, val_time_augmentation=None, test_time_augmentation=None, update_bn_momentum=False, freeze_bn_layers=False)[source]¶

Fine-tuning training.

Parameters:

trunk (DictConfig) – Config to build a trunk.
classifier (DictConfig) – Config to build a classifier.
optimizer (DictConfig) – Config to build an optimizer for trunk.
pretrained_trunk_path (str) – Path to the pretrained trunk file.
trunk_pattern (str, optional) – Pattern to retrieve the trunk model in checkpoint state_dict and delete the key.
Default: '^(trunk\.)'
two_groups (bool, optional) – If True, use two groups of parameters for optimizer, the trunk and the head.
Default: False
freeze_trunk (bool, optional) – If True, freeze the trunk.
Default: False
train_transform (Optional[DictConfig], optional) – Config to perform transformation on train input.
Default: None
val_transform (Optional[DictConfig], optional) – Config to perform transformation on val input.
Default: None
test_transform (Optional[DictConfig], optional) – Config to perform transformation on test input.
Default: None
val_time_augmentation (Optional[DictConfig], optional) – Ensembling method for test time augmentation used at validation.
Default: None
test_time_augmentation (Optional[DictConfig], optional) – Ensembling method for test time augmentation used at test.
Default: None
update_bn_momentum (bool, optional) – If True update batch norm statistics according to \(max(1 - 10/steps\_per\_epoch, 0.9)\).
Default: False
freeze_bn_layers (bool, optional) – If True, freeze the batch norm layers.
Default: False

Example:

trunk = {...} # config to build a trunk
classifier = {...} # config to build a classifier
optimizer = {...} # config to build an optimizer
pretrained_trunk_path = ... # path where the trunk has been saved

model = FinetuningModel(trunk, classifier, optimizer, pretrained_trunk_path)

class eztorch.models.soccernet_spotting.SoccerNetSpottingModel(trunk, head_class, optimizer, pretrained_path=None, pretrained_trunk_path=None, loss_fn_args={}, prediction_args={}, eval_step_timestamp=1.0, trunk_pattern='^(trunk\\\\.)', freeze_trunk=False, train_transform=None, val_transform=None, test_transform=None, save_val_preds_path='val_preds/', save_test_preds_path='test_preds/', NMS_args={'window': 10, 'threshold': 0.001}, evaluation_args={}, do_compile=False)[source]¶

Model to perform action spotting.

Parameters:

trunk (DictConfig) – Config to build a trunk.
head_class (DictConfig) – Config to build a head for classification.
optimizer (DictConfig) – Config to build an optimizer for trunk.
pretrained_trunk_path (str | None, optional) – Path to the pretrained trunk file.
Default: None
pretrained_path (str | None, optional) – Path to the pretrained model.
Default: None
prediction_args (DictConfig, optional) – Arguments to configure predictions.
Default: {}
loss_fn_args (DictConfig, optional) – Arguments for the loss function.
Default: {}
eval_step_timestamp (float, optional) – Step between each timestamp.
Default: 1.0
trunk_pattern (str, optional) – Pattern to retrieve the trunk model in checkpoint state_dict and delete the key.
Default: '^(trunk\.)'
freeze_trunk (bool, optional) – Whether to freeze the trunk.
Default: False
train_transform (DictConfig | None, optional) – Config to perform transformation on train input.
Default: None
val_transform (DictConfig | None, optional) – Config to perform transformation on val input.
Default: None
test_transform (DictConfig | None, optional) – Config to perform transformation on test input.
Default: None
save_val_preds_path (str | Path, optional) – Path to store the validation predictions.
Default: 'val_preds/'
save_test_preds_path (str | Path, optional) – Path to store the test predictions.
Default: 'test_preds/'
NMS_args (DictConfig, optional) – Arguments to configure the NMS.
Default: {'window': 10, 'threshold': 0.001}
evaluation_args (DictConfig, optional) – Arguments to configure the evaluation.
Default: {}

Example:

trunk = {...} # config to build a trunk
head_class = {...} # config to build a head for classification
optimizer = {...} # config to build an optimizer
pretrained_trunk_path = ... # path where the trunk has been saved

model = SoccerNetSpottingModel(trunk, head_class, optimizer, pretrained_trunk_path)

class eztorch.models.spotting.SpottingModel(trunk, head_class, optimizer, pretrained_path=None, pretrained_trunk_path=None, loss_fn_args={}, prediction_args={}, trunk_pattern='^(trunk\\\\.)', freeze_trunk=False, train_transform=None, val_transform=None, test_transform=None, save_val_preds_path='val_preds/', save_test_preds_path='test_preds/', NMS_args={'window': 10, 'threshold': 0.001}, evaluation_args={}, do_compile=False)[source]¶

Model to perform spotting.

Parameters:

trunk (DictConfig) – Config to build a trunk.
head_class (DictConfig) – Config to build a head for classification.
optimizer (DictConfig) – Config to build an optimizer for trunk.
pretrained_trunk_path (str | None, optional) – Path to the pretrained trunk file.
Default: None
pretrained_path (str | None, optional) – Path to the pretrained model.
Default: None
prediction_args (DictConfig, optional) – Arguments to configure predictions.
Default: {}
loss_fn_args (DictConfig, optional) – Arguments for the loss function.
Default: {}
trunk_pattern (str, optional) – Pattern to retrieve the trunk model in checkpoint state_dict and delete the key.
Default: '^(trunk\.)'
freeze_trunk (bool, optional) – Whether to freeze the trunk.
Default: False
train_transform (DictConfig | None, optional) – Config to perform transformation on train input.
Default: None
val_transform (DictConfig | None, optional) – Config to perform transformation on val input.
Default: None
test_transform (DictConfig | None, optional) – Config to perform transformation on test input.
Default: None
save_val_preds_path (str | Path, optional) – Path to store the validation predictions.
Default: 'val_preds/'
save_test_preds_path (str | Path, optional) – Path to store the test predictions.
Default: 'test_preds/'
NMS_args (DictConfig, optional) – Arguments to configure the NMS.
Default: {'window': 10, 'threshold': 0.001}
evaluation_args (DictConfig, optional) – Arguments to configure the evaluation.
Default: {}

Example:

trunk = {...} # config to build a trunk
head_class = {...} # config to build a head for classification
optimizer = {...} # config to build an optimizer
pretrained_trunk_path = ... # path where the trunk has been saved

model = SpottingModel(trunk, head_class, optimizer, pretrained_trunk_path)