def main(args):
logger.info('Using torch {}'.format(torch.__version__))
cfg = CommonConfiguration.from_yaml(args.setting)
logger.info('Loaded training setting file: {}'.format(args.setting))
trainer = Trainer(cfg)
logger.info('Started training...')
trainer.start()
logger.info('Finished training.')
def _parser_dict(self):
dictionary = CommonConfiguration.from_yaml(cfg.DATASET.DICTIONARY)
return dictionary[cfg.DATASET.DICTIONARY_NAME]
perf_log += "{:}: {:.4f}\n".format(k, v)
perf_log += "------------------------------------\n"
logger.info(perf_log)
acc = performances['performance']
return acc, perf_log
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Generic Pytorch-based Training Framework')
parser.add_argument('--setting', default='conf/cityscapes_deeplabv3plus.yml', help='The path to the configuration file.')
# distributed training parameters
parser.add_argument("--local_rank", default=0, type=int)
args = parser.parse_args()
cfg = CommonConfiguration.from_yaml(args.setting)
cfg.local_rank = args.local_rank
if cfg.local_rank==0:
logger.info('Loaded configuration file: {}'.format(args.setting))
logger.info('Use gpu ids: {}'.format(cfg.GPU_IDS))
trainer = Trainer(cfg)
logger.info('Begin to training ...')
trainer.run()
if cfg.local_rank == 0:
logger.info('finish!')
torch.cuda.empty_cache()
def _parser_dict(self):
dictionary = CommonConfiguration.from_yaml(cfg.DATASET.DICTIONARY)
return next(dictionary.items())[1] ## return first