def _generate_prediction(meta_data, pipeline, logger, category_ids):
data = {'input': {'meta': meta_data,
'target_sizes': [(300, 300)] * len(meta_data),
},
'specs': {'train_mode': True},
'callback_input': {'meta_valid': None}
}
pipeline.clean_cache()
output = pipeline.transform(data)
pipeline.clean_cache()
y_pred = output['y_pred']
prediction = create_annotations(meta_data, y_pred, logger, category_ids)
return prediction
def _generate_prediction_in_chunks(meta_data, pipeline, logger, category_ids, chunk_size):
prediction = []
for meta_chunk in generate_data_frame_chunks(meta_data, chunk_size):
data = {'input': {'meta': meta_chunk,
'target_sizes': [(300, 300)] * len(meta_chunk)
},
'specs': {'train_mode': True},
'callback_input': {'meta_valid': None}
}
pipeline.clean_cache()
output = pipeline.transform(data)
pipeline.clean_cache()
y_pred = output['y_pred']
prediction_chunk = create_annotations(meta_chunk, y_pred, logger, category_ids)
prediction.extend(prediction_chunk)
return prediction
def _generate_prediction(self, cache_dirpath, outputs):
data = {
'callback_input': {
'meta': self.meta_valid,
'meta_valid': None,
'target_sizes': [(300, 300)] * len(self.meta_valid),
},
'unet_output': {
**outputs
}
}
pipeline = self.validation_pipeline(cache_dirpath)
for step_name in pipeline.all_steps:
cmd = 'touch {}'.format(
os.path.join(cache_dirpath, 'transformers', step_name))
subprocess.call(cmd, shell=True)
output = pipeline.transform(data)
y_pred = output['y_pred']
prediction = create_annotations(self.meta_valid, y_pred, logger,
CATEGORY_IDS)
return prediction