fr3dnet_trainer.py 文件源码

def __init__(self):
        metric_names = ['Loss','L2','Accuracy']
        super(Fr3dNetTrainer, self).__init__(metric_names)

        tensor5 = T.TensorType(theano.config.floatX, (False,) * 5)
        input_var = tensor5('inputs')
        target_var = T.ivector('targets')

        logging.info("Defining network")
        net = fr3dnet.define_network(input_var)
        self.network = net
        train_fn, val_fn, l_r = fr3dnet.define_updates(net, input_var, target_var)

        self.train_fn = train_fn
        self.val_fn = val_fn
        self.l_r = l_r