addcb13fb3dcd2cdbc32ac81611a78ac1c19b02f,niftynet/layer/channel_sparse_convolution.py,ChannelSparseConvLayer,layer_op,#ChannelSparseConvLayer#Any#Any#Any#,106

Before Change

    n_full_input_chns = _input_mask.get_shape().as_list()[0]
    spatial_rank = layer_util.infer_spatial_rank(input_tensor)
    // initialize conv kernels/strides and then apply
    w_full_size = np.vstack((
        [self.kernel_size] * spatial_rank,
        n_full_input_chns, self.n_output_chns)).flatten()
    full_stride = np.vstack((
        [self.stride] * spatial_rank)).flatten()
    conv_kernel = tf.get_variable(
        "w", shape=w_full_size.tolist(),
        initializer=self.initializers["w"],

After Change

    n_full_input_chns = _input_mask.get_shape().as_list()[0]
    spatial_rank = layer_util.infer_spatial_rank(input_tensor)
    // initialize conv kernels/strides and then apply
    w_full_size = layer_util.expand_spatial_params(
        self.kernel_size, spatial_rank)
    // expand kernel size to include number of features
    w_full_size = w_full_size + (n_full_input_chns, self.n_output_chns)

    full_stride = layer_util.expand_spatial_params(
        self.stride, spatial_rank)

    full_dilation = layer_util.expand_spatial_params(
        self.dilation, spatial_rank)

    conv_kernel = tf.get_variable(
        "w", shape=w_full_size,

Instances