easy_vision.python.main

Binary to run train and evaluation on easy vision model.

easy_vision.python.main.train_and_evaluate(pipeline_config_path, continue_train=False)

Build an CVEstimator, and then train and evaluate the estimator.

Parameters:

• pipeline_config_path – a path to proto.CvEstimator object, specifies
• train_config – model_config, data_config and eval_config
• continue_train – whether to restart train from an existing checkpoint

Returns:

None, the model will be saved into train_config.model_dir

easy_vision.python.main.train_and_evaluate_with_param_config(param_config_str, continue_train=False, data_prefix='')

train and evaluate the model with parameters.

Parameters:

• param_config_str – model parameters config string
• continue_train – whether to restart train from an existing checkpoint
• data_prefix – data path prefix for data and pretrained-models, e.g., osspath oss://pai-vision-data/, local path /home/user/data

Returns:

None, the model will be saved into train_config.model_dir

easy_vision.python.main.evaluate(pipeline_config_path, eval_checkpoint_path='', eval_data_path=None, eval_result_filename='eval_result.txt')

Evaluate for evaluation for eval data in pipeline_config_path, the metrics will also be displayed on tensorboard.

Parameters:

• pipeline_config_path – config for the model, eval_data, eval_config
• eval_checkpoint_path – if specified, will use this model instead of model specified by model_dir in pipeline_config_path
• eval_data_path – eval data path, default use eval data in pipeline_config could be a path or a list of paths

Returns:

a dict of evaluation metrics, the metrics are specified in

pipeline_config_path

Raises:

AssertionError – if pipeline_config_path does not exist

easy_vision.python.main.predict(pipeline_config_path, test_checkpoint_path='', test_filelist=None)

Test for evaluation for eval_data in pipeline_config_path

Parameters:

• pipeline_config_path – file specify proto.CvEstimator, including model_config, eval_data, eval_config
• test_checkpoint_path – if specified, will use this model instead of model in model_dir in pipeline_config_path

Returns:

a list of items, each item represent prediction result for one image

Raises:

AssertionError – (a) if pipeline_config_path does not exist. (b) if train_config.model_dir does not exist.

easy_vision.python.main.export(export_dir, pipeline_config_path, checkpoint_path='')

Export model defined in pipeline_config_path

Parameters:

• export_dir – base directory where the model should be exported
• pipeline_config_path – file specify proto.CvEstimator, including model_config, eval_data, eval_config
• checkpoint_path – if specified, will use this model instead of model in model_dir in pipeline_config_path

Returns:

the directory where model is exported

Raises:

AssertionError – if pipeline_config_path does not exist

easy_vision.python.main.predictor_evaluate(config_path)

Evaluator a predictor

Parameters:config_path – a path to proto.PredictorEval object Returns:a dict of evaluation metrics Raises:AssertionError – if config_path does not exist