objectscope.trainer

Classes

TrainSession

Module Contents

class objectscope.trainer.TrainSession(train_img_dir, train_coco_json_file, test_img_dir, test_coco_json_file, config_file_url, num_classes, train_data_name=None, test_data_name=None, train_metadata={}, test_metadata={}, output_dir='output/object_detector', device='cuda', num_workers=12, imgs_per_batch=4, base_lr=5e-05, max_iter=5000, checkpoint_period=50, **kwargs)

Bases: object

train_img_dir
train_coco_json_file
test_img_dir
test_coco_json_file
train_metadata
test_metadata
config_file_url
num_classes
output_dir = 'output/object_detector'
device = 'cuda'
num_workers = 12
imgs_per_batch = 4
base_lr = 5e-05
max_iter = 5000
checkpoint_period = 50
test_data_name = None
train_data_name = None
register_dataset(train_img_dir=None, train_coco_json_file=None, test_img_dir=None, test_coco_json_file=None, train_data_name=None, test_data_name=None, train_metadata={}, test_metadata={})
create_config(num_classes=None, config_file_url=None, num_workers=None, imgs_per_batch=None, base_lr=None, max_iter=None, checkpoint_period=None, output_dir=None, device=None, train_data_name=None, test_data_name=None, anchor_ratios: None | List[List] = None, anchor_sizes: None | List[List] = None, evaluate_period=1)

_summary_

Parameters:

  • num_classes (_type_, optional) – _description_. Defaults to None.

  • config_file_url (_type_, optional) – _description_. Defaults to None.

  • num_workers (_type_, optional) – _description_. Defaults to None.

  • imgs_per_batch (_type_, optional) – _description_. Defaults to None.

  • base_lr (_type_, optional) – _description_. Defaults to None.

  • max_iter (_type_, optional) – _description_. Defaults to None.

  • checkpoint_period (_type_, optional) – _description_. Defaults to None.

  • output_dir (_type_, optional) – _description_. Defaults to None.

  • device (_type_, optional) – _description_. Defaults to None.

  • train_data_name (_type_, optional) – _description_. Defaults to None.

  • test_data_name (_type_, optional) – _description_. Defaults to None.

  • anchor_ratios (Union[None, List[List]], optional) – Anchor ratios use for generating anchor boxes. Example [[0.7685566328549631, 1.8715268243900367, 1.1942387054643602]].

  • anchor_sizes (list, optional) – Anchor sizes use for generating anchor boxes and RPN. Example [[240.9236833908755], [59.864835712691715], [153.60699447681742], [434.33823627084996], [103.37411650130916]].

  • evaluate_period (int, optional) – _description_. Defaults to 1.

Returns:

_description_

Return type:

_type_

get_trainer(cfg=None)
create_trainer()
run()