def get_composed_augmentations(aug_dict):
if aug_dict is None:
logger.info("Using No Augmentations")
return None
augmentations = []
for aug_key, aug_param in aug_dict.items():
augmentations.append(key2aug[aug_key](aug_param))
logger.info("Using {} aug with params {}".format(aug_key, aug_param))
return Compose(augmentations)
def inference(cfg, model, val_loader):
logger.info("Start infer")
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu'
) if cfg.MODEL.DEVICE == 'cuda' else 'cpu'
evaluator = create_supervised_evaluator(model,
metrics={'accuracy': Accuracy()},
device=device)
# adding handlers using `evaluator.on` decorator API
@evaluator.on(Events.EPOCH_COMPLETED)
def print_validation_results(engine):
metrics = evaluator.state.metrics
avg_acc = metrics['accuracy']
logger.info("Validation Results - Accuracy: {:.3f}".format(avg_acc))
evaluator.run(val_loader)
def print_validation_results(engine):
metrics = evaluator.state.metrics
avg_acc = metrics['accuracy']
logger.info("Validation Results - Accuracy: {:.3f}".format(avg_acc))
from ignite.engine import Events
from ignite.engine import create_supervised_evaluator
from ignite.metrics import Accuracy
sys.path.append('../..')
from cvnet.utils.logger import logger
def inference(cfg, model, val_loader):
logger.info("Start infer")
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu'
) if cfg.MODEL.DEVICE == 'cuda' else 'cpu'
evaluator = create_supervised_evaluator(model,
metrics={'accuracy': Accuracy()},
device=device)
# adding handlers using `evaluator.on` decorator API
@evaluator.on(Events.EPOCH_COMPLETED)
def print_validation_results(engine):
metrics = evaluator.state.metrics
avg_acc = metrics['accuracy']
logger.info("Validation Results - Accuracy: {:.3f}".format(avg_acc))
evaluator.run(val_loader)
if __name__ == '__main__':
logger.info('infer')
