'min_kspace_width': None,
'max_kspace_width': 740,
'min_kspace_height': None,
'max_kspace_height': None,
'acquisition_type': None,
'system_type': None,
'min_target_width': None,
'min_target_height': None,
'only_square_targets': True,
'scale_inputs': False,
'num_coils': 15,
'magnet': 1.5,
'accelerations': [4],
'num_low_frequencies': [16],
'train_accelerations': [4],
'train_num_low_frequencies': [16],
'initialization': 'none',
'log_interval': 10,
'visual_first_epoch': False,
# After pretraining, link to pretrained model file here
#'checkpoint_type': "restart",
#'checkpoint': 'pretrained_orien_pat2_10x.mdl',
'orientation_augmentation': True,
'orientation_adversary': True,
}
if __name__ == "__main__":
#spawn_dist.run(config) # Multiple GPU training (8 recommended)
run.run(config) # Single GPU training
'calculate_offsets_directly': True,
'workers': 0,
'filter_acceleration': 1,
'min_kspace_width': None,
'max_kspace_width': 740,
'min_kspace_height': None,
'max_kspace_height': None,
'acquisition_type': None,
'system_type': None,
'min_target_width': None,
'min_target_height': None,
'only_square_targets': True,
'scale_inputs': False,
'num_coils': 15,
'magnet': 1.5,
'accelerations': [4],
'num_low_frequencies': [16],
'train_accelerations': [4],
'train_num_low_frequencies': [16],
'initialization': 'none',
'sync_params': False,
'log_interval': 10,
'visual_first_epoch': False,
'orientation_augmentation': True,
'orientation_adversary': False,
}
if __name__ == "__main__":
#spawn_dist.run(config) # Multiple GPU training (8 recommended)
run.run(config) # Single GPU training (for debugging)
'magnet': 1.5,
'accelerations': [4],
'num_low_frequencies': [16],
'train_accelerations': [4],
'train_num_low_frequencies': [16],
'initialization': 'none',
'log_interval': 10,
'visual_first_epoch': False,
'display_count': 8,
# After pretraining, link to pretrained model file here
'checkpoint_type': "restart",
'checkpoint': f'{pathlib.Path.home()}/pretrained_lowt_4x.mdl',
'orientation_augmentation': True,
'orientation_adversary': True,
'adversary_model': 'shallow',
'adversary_epoch_from': 0,
'warmup_adversary_from': 0,
'adversary_strength': 1,
'reg_param': 0.1,
'adversary_weight_decay': 0.0,
'ssim_l1_coefficient': 0.01,
'number_of_adversaries': 1,
'evaluate': False,
'short_epochs': True, # 20% epochs to avoid memory issues
'epochs': 1000,
}
if __name__ == "__main__":
run.run(config) # Single GPU training for debugging only