d0af71e1ea3c1f0de2bc229b623c77b991fac64d,layer/highres3dnet.py,HighRes3DNet,layer_op,#HighRes3DNet#Any#Any#Any#,44

Before Change

        //// convolutions  dilation factor = 2
        res_out = tf.space_to_batch_nd(res_out, [2, 2, 2], zero_paddings)
        for i in range(self.num_res_blocks[1]):
            res_block = HighResBlock(self.num_features[1], kernels=(3, 3), with_res=True, name="res_2_{}".format(i))
            res_out = res_block(res_out, is_training)
        res_out = tf.batch_to_space_nd(res_out, [2, 2, 2], zero_paddings)

        //// convolutions  dilation factor = 4
        res_out = tf.space_to_batch_nd(res_out, [4, 4, 4], zero_paddings)
        for i in range(self.num_res_blocks[2]):
            res_block = HighResBlock(self.num_features[2], kernels=(3, 3), with_res=True, name="res_3_{}".format(i))
            res_out = res_block(res_out, is_training)
        res_out = tf.batch_to_space_nd(res_out, [4, 4, 4], zero_paddings)

After Change

        //// convolutions  dilation factor = 2
        with DilatedTensor(res_out, 2) as dilated:
            for i in range(self.num_res_blocks[1]):
                res_block = HighResBlock(self.num_features[1],
                                         kernels=(3, 3),
                                         with_res=True,
                                         name="res_1_{}".format(i))
                dilated.tensor = res_block(dilated.tensor, is_training)
        res_out = dilated.tensor

        //// convolutions  dilation factor = 4
        with DilatedTensor(res_out, 4) as dilated:
            for i in range(self.num_res_blocks[2]):
                res_block = HighResBlock(self.num_features[2],
                                         kernels=(3, 3),
                                         with_res=True,
                                         name="res_2_{}".format(i))
                dilated.tensor = res_block(dilated.tensor, is_training)
        res_out = dilated.tensor

        //// 1x1x1 convolution "fully connected"
        conv_kernel_1_op = ConvolutionalLayer(
                self.num_features[3], kernel_size=1, name="con_fc_1")

Instances