def __init__(self, config):
"""Initialize configurations."""
self.image_size = config['image_size']
self.class_num = config['class_num']
self.class_names = config['class_names']
self.k_proposals = config['k_proposals']
self.balance_factor = config['balance_factor']
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.basebone = fc_resnet50(self.class_num, True)
self.prm_module = peak_response_mapping(self.basebone, **config['model'])
self.filling_module = instance_extent_filling(self.basebone, config)
self.cuda = torch.cuda.is_available()
if self.cuda:
self.prm_module.to(self.device)
self.filling_module.to(self.device)
self.prm_module_criterion = multilabel_soft_margin_loss
self.filling_module_criterion = binary_cross_entropy_loss
self.max_epoch = config['max_epoch']
self.params = finetune(self.prm_module, **config['finetune'])
self.optimizer_prm = sgd_optimizer(self.params, **config['optimizer'])
self.params = finetune(self.filling_module, **config['finetune'])
self.optimizer_filling = sgd_optimizer(self.params, **config['optimizer'])
self.lr_update_step = 999999
self.lr = config['optimizer']['lr']
self.snapshot = config['snapshot']
def __init__(self, config):
"""Initialize configurations."""
self.basebone = fc_resnet50(20, True)
self.model = peak_response_mapping(self.basebone, **config['model'])
self.criterion = multilabel_soft_margin_loss
self.max_epoch = config['max_epoch']
self.cuda = (config['device'] == 'cuda')
self.params = finetune(self.model, **config['finetune'])
# print(self.params)
self.optimizer = sgd_optimizer(self.params, **config['optimizer'])
self.lr_update_step = 999999
self.lr = config['optimizer']['lr']
self.snapshot = config['snapshot']
if self.cuda:
self.model.to('cuda')