Python mxnet.nd.split() Examples

def forward(self, x):
        '''
        Parameters
        ----------
        x: nd.array, shape is (batch_size, c_in, time_step, num_of_vertices)

        Returns
        ----------
        shape is (batch_size, c_out, time_step - Kt + 1, num_of_vertices)

        '''

        x_input = self.align(x)[:, :, self.Kt - 1:, :]

        x_conv = self.conv(x)
        if self.activation == 'GLU':
            x_conv = self.conv(x)
            x_conv1, x_conv2 = nd.split(x_conv, axis=1, num_outputs=2)
            return (x_conv1 + x_input) * nd.sigmoid(x_conv2)
        if self.activation == 'relu':
            return nd.relu(x_conv + x_input)
        return x_conv 

def forward(self, samples, matches, anchors, refs):
        """Forward"""
        F = nd
        # TODO(zhreshold): batch_pick, take multiple elements?
        ref_boxes = nd.repeat(refs.reshape((0, 1, -1, 4)), axis=1, repeats=matches.shape[1])
        ref_boxes = nd.split(ref_boxes, axis=-1, num_outputs=4, squeeze_axis=True)
        ref_boxes = nd.concat(*[F.pick(ref_boxes[i], matches, axis=2).reshape((0, -1, 1)) \
            for i in range(4)], dim=2)
        g = self.corner_to_center(ref_boxes)
        a = self.corner_to_center(anchors)
        t0 = ((g[0] - a[0]) / a[2] - self._means[0]) / self._stds[0]
        t1 = ((g[1] - a[1]) / a[3] - self._means[1]) / self._stds[1]
        t2 = (F.log(g[2] / a[2]) - self._means[2]) / self._stds[2]
        t3 = (F.log(g[3] / a[3]) - self._means[3]) / self._stds[3]
        codecs = F.concat(t0, t1, t2, t3, dim=2)
        temp = F.tile(samples.reshape((0, -1, 1)), reps=(1, 1, 4)) > 0.5
        targets = F.where(temp, codecs, F.zeros_like(codecs))
        masks = F.where(temp, F.ones_like(temp), F.zeros_like(temp))
        return targets, masks 

def hybrid_forward(self, F, x, anchors):
        if self.corner_to_center is not None:
            a = self.corner_to_center(anchors)
        else:
            a = anchors.split(axis=-1, num_outputs=4)
        p = F.split(x, axis=-1, num_outputs=4)
        ox = F.broadcast_add(F.broadcast_mul(p[0] * self._stds[0] + self._means[0], a[2]), a[0])
        oy = F.broadcast_add(F.broadcast_mul(p[1] * self._stds[1] + self._means[1], a[3]), a[1])
        tw = F.exp(p[2] * self._stds[2] + self._means[2])
        th = F.exp(p[3] * self._stds[3] + self._means[2])
        if self._clip:
            tw = F.minimum(tw, self._clip)
            th = F.minimum(th, self._clip)
        ow = F.broadcast_mul(tw, a[2]) / 2
        oh = F.broadcast_mul(th, a[3]) / 2
        return F.concat(ox - ow, oy - oh, ox + ow, oy + oh, dim=-1) 

def extract_multi_position_matrix_nd(bbox):
    bbox = nd.transpose(bbox, axes=(1, 0, 2))
    xmin, ymin, xmax, ymax = nd.split(data=bbox, num_outputs=4, axis=2)
    # [num_fg_classes, num_boxes, 1]
    bbox_width = xmax - xmin + 1.
    bbox_height = ymax - ymin + 1.
    center_x = 0.5 * (xmin + xmax)
    center_y = 0.5 * (ymin + ymax)
    # [num_fg_classes, num_boxes, num_boxes]
    delta_x = nd.broadcast_minus(lhs=center_x, 
        rhs=nd.transpose(center_x, axes=(0, 2, 1)))
    delta_x = nd.broadcast_div(delta_x, bbox_width)
    delta_x = nd.log(nd.maximum(nd.abs(delta_x), 1e-3))

    delta_y = nd.broadcast_minus(lhs=center_y,
        rhs=nd.transpose(center_y, axes=(0, 2, 1)))
    delta_y = nd.broadcast_div(delta_y, bbox_height)
    delta_y = nd.log(nd.maximum(nd.abs(delta_y), 1e-3))

    delta_width = nd.broadcast_div(lhs=bbox_width, 
        rhs=nd.transpose(bbox_width, axes=(0, 2, 1)))
    delta_width = nd.log(delta_width)

    delta_height = nd.broadcast_div(lhs=bbox_height,
        rhs=nd.transpose(bbox_height, axes=(0, 2, 1)))
    delta_height = nd.log(delta_height)
    concat_list = [delta_x, delta_y, delta_width, delta_height]
    for idx, sym in enumerate(concat_list):
        concat_list[idx] = nd.expand_dims(sym, axis=3)
    position_matrix = nd.concat(*concat_list, dim=3)
    return position_matrix 

def forward(self, feature, data, begin_state):
        num_nodes, batch_size, length, _ = data.shape

        data = nd.split(data, axis=2, num_outputs=length, squeeze_axis=1)

        outputs, state = [], begin_state
        for input in data:
            output, state = self.forward_single(feature, input, state)
            outputs.append(output)

        outputs = nd.stack(*outputs, axis=2)
        return outputs, state 

def forward(self, feature, data, begin_state):
        num_nodes, batch_size, length, _ = data.shape

        data = nd.split(data, axis=2, num_outputs=length, squeeze_axis=1)

        outputs, state = [], begin_state
        for input in data:
            output, state = self.forward_single(feature, input, state)
            outputs.append(output)

        outputs = nd.stack(*outputs, axis=2)
        return outputs, state 

def extract_multi_position_matrix_nd(bbox):
    bbox = nd.transpose(bbox, axes=(1, 0, 2))
    xmin, ymin, xmax, ymax = nd.split(data=bbox, num_outputs=4, axis=2)
    # [num_fg_classes, num_boxes, 1]
    bbox_width = xmax - xmin + 1.
    bbox_height = ymax - ymin + 1.
    center_x = 0.5 * (xmin + xmax)
    center_y = 0.5 * (ymin + ymax)
    # [num_fg_classes, num_boxes, num_boxes]
    delta_x = nd.broadcast_minus(lhs=center_x, 
        rhs=nd.transpose(center_x, axes=(0, 2, 1)))
    delta_x = nd.broadcast_div(delta_x, bbox_width)
    delta_x = nd.log(nd.maximum(nd.abs(delta_x), 1e-3))

    delta_y = nd.broadcast_minus(lhs=center_y,
        rhs=nd.transpose(center_y, axes=(0, 2, 1)))
    delta_y = nd.broadcast_div(delta_y, bbox_height)
    delta_y = nd.log(nd.maximum(nd.abs(delta_y), 1e-3))

    delta_width = nd.broadcast_div(lhs=bbox_width, 
        rhs=nd.transpose(bbox_width, axes=(0, 2, 1)))
    delta_width = nd.log(delta_width)

    delta_height = nd.broadcast_div(lhs=bbox_height,
        rhs=nd.transpose(bbox_height, axes=(0, 2, 1)))
    delta_height = nd.log(delta_height)
    concat_list = [delta_x, delta_y, delta_width, delta_height]
    for idx, sym in enumerate(concat_list):
        concat_list[idx] = nd.expand_dims(sym, axis=3)
    position_matrix = nd.concat(*concat_list, dim=3)
    return position_matrix 

def __init__(self, stds=(0.1, 0.1, 0.2, 0.2), means=(0., 0., 0., 0.)):
        super(NormalizedBoxCenterEncoder, self).__init__()
        assert len(stds) == 4, "Box Encoder requires 4 std values."
        self._stds = stds
        self._means = means
        with self.name_scope():
            self.corner_to_center = BBoxCornerToCenter(split=True) 

def __init__(self, num_class, stds=(0.1, 0.1, 0.2, 0.2), means=(0., 0., 0., 0.)):
        super(NormalizedPerClassBoxCenterEncoder, self).__init__()
        assert len(stds) == 4, "Box Encoder requires 4 std values."
        assert num_class > 0, "Number of classes must be positive"
        self._num_class = num_class
        self._stds = stds
        self._means = means
        with self.name_scope():
            self.corner_to_center = BBoxCornerToCenter(split=True) 

def forward(self, samples, matches, anchors, labels, refs):
        """Encode BBox One entry per category"""
        F = nd
        ref_boxes = F.repeat(refs.reshape((0, 1, -1, 4)), axis=1, repeats=matches.shape[1])
        ref_boxes = F.split(ref_boxes, axis=-1, num_outputs=4, squeeze_axis=True)
        ref_boxes = F.concat(*[F.pick(ref_boxes[i], matches, axis=2).reshape((0, -1, 1)) \
            for i in range(4)], dim=2)
        ref_labels = F.repeat(labels.reshape((0, 1, -1)), axis=1, repeats=matches.shape[1])
        ref_labels = F.pick(ref_labels, matches, axis=2).reshape((0, -1, 1))
        g = self.corner_to_center(ref_boxes)
        a = self.corner_to_center(anchors)
        t0 = ((g[0] - a[0]) / a[2] - self._means[0]) / self._stds[0]
        t1 = ((g[1] - a[1]) / a[3] - self._means[1]) / self._stds[1]
        t2 = (F.log(g[2] / a[2]) - self._means[2]) / self._stds[2]
        t3 = (F.log(g[3] / a[3]) - self._means[3]) / self._stds[3]
        codecs = F.concat(t0, t1, t2, t3, dim=2)
        temp = F.tile(samples.reshape((0, -1, 1)), reps=(1, 1, 4)) > 0.5
        targets = F.where(temp, codecs, F.zeros_like(codecs))
        masks = F.where(temp, F.ones_like(temp), F.zeros_like(temp))
        out_targets = []
        out_masks = []
        for cid in range(self._num_class):
            same_cid = ref_labels == cid
            # keep orig targets
            out_targets.append(targets)
            # but mask out the one not belong to this class
            out_masks.append(masks * same_cid.repeat(axis=-1, repeats=4))
        all_targets = F.stack(*out_targets, axis=0)
        all_masks = F.stack(*out_masks, axis=0)
        return all_targets, all_masks 

def __init__(self, num_class, stds=(0.1, 0.1, 0.2, 0.2), means=(0., 0., 0., 0.)):
        super(NormalizedBoxCenterEncoder_Cascade, self).__init__()
        assert len(stds) == 4, "Box Encoder requires 4 std values."
        assert num_class > 0, "Number of classes must be positive"
        self._num_class = num_class
        self._stds = stds
        self._means = means
        with self.name_scope():
            self.corner_to_center = BBoxCornerToCenter(split=True) 

def __init__(self, stds=(0.1, 0.1, 0.2, 0.2), means=(0., 0., 0., 0.),
                 convert_anchor=False, clip=None):
        super(NormalizedBoxCenterDecoder, self).__init__()
        assert len(stds) == 4, "Box Encoder requires 4 std values."
        self._stds = stds
        self._means = means
        self._clip = clip
        if convert_anchor:
            self.corner_to_center = BBoxCornerToCenter(split=True)
        else:
            self.corner_to_center = None 

def refine_bbox_nd(bbox, bbox_delta, im_info=None, means=None, stds=None):

    xmin, ymin, xmax, ymax = nd.split(data=bbox, num_outputs=4, axis=1)
    bbox_width = xmax - xmin + 1.
    bbox_height = ymax - ymin + 1.
    center_x = 0.5 * (xmin + xmax)
    center_y = 0.5 * (ymin + ymax)

    bbox_delta_reshape = nd.Reshape(data=bbox_delta, shape=(0, -1, 4))
    dx, dy, dw, dh = nd.split(data=bbox_delta_reshape, 
        num_outputs=4, axis=2, squeeze_axis=1)
    if (means is not None) and (stds is not None):
        dx = dx * stds[0] + means[0]
        dy = dy * stds[1] + means[1]
        dw = dw * stds[2] + means[2]
        dh = dh * stds[3] + means[3]

    refine_center_x = nd.broadcast_add(lhs=center_x,
        rhs=nd.broadcast_mul(lhs=bbox_width, rhs=dx))
    refine_center_y = nd.broadcast_add(lhs=center_y,
        rhs=nd.broadcast_mul(lhs=bbox_height, rhs=dy))
    refined_width = nd.broadcast_mul(lhs=bbox_width, rhs=nd.exp(dw))
    refined_height = nd.broadcast_mul(lhs=bbox_height, rhs=nd.exp(dh))
    w_offset = 0.5 * (refined_width - 1.)
    h_offset = 0.5 * (refined_height - 1.)
    refined_xmin = nd.expand_dims(refine_center_x - w_offset, axis=1)
    refined_ymin = nd.expand_dims(refine_center_y - h_offset, axis=1)
    refined_xmax = nd.expand_dims(refine_center_x + w_offset, axis=1)
    refined_ymax = nd.expand_dims(refine_center_y + h_offset, axis=1)

    refined_bbox = nd.concat(refined_xmin, refined_ymin, refined_xmax, refined_ymax, dim=1)
    if im_info is not None:
        # assume im_info [[height, width, scale]] with shape (1,3)
        im_hw = nd.slice_axis(im_info, axis=1, begin=0, end=2)
        im_wh = nd.reverse(im_hw, axis=1)
        im_wh = im_wh - 1.
        im_wh = nd.tile(data=im_wh, reps=(1, 2))
        im_wh = nd.Reshape(im_wh, shape=(1, 4, 1))
        refined_bbox = nd.broadcast_minimum(lhs=refined_bbox, rhs=im_wh)
        refined_bbox = nd.broadcast_maximum(lhs=refined_bbox,
            rhs=nd.zeros_like(refined_bbox))
    # print refined_bbox.debug_str()
    return refined_bbox 

def refine_bbox_nd(bbox, bbox_delta, im_info=None, means=None, stds=None):

    xmin, ymin, xmax, ymax = nd.split(data=bbox, num_outputs=4, axis=1)
    bbox_width = xmax - xmin + 1.
    bbox_height = ymax - ymin + 1.
    center_x = 0.5 * (xmin + xmax)
    center_y = 0.5 * (ymin + ymax)

    bbox_delta_reshape = nd.Reshape(data=bbox_delta, shape=(0, -1, 4))
    dx, dy, dw, dh = nd.split(data=bbox_delta_reshape, 
        num_outputs=4, axis=2, squeeze_axis=1)
    if (means is not None) and (stds is not None):
        dx = dx * stds[0] + means[0]
        dy = dy * stds[1] + means[1]
        dw = dw * stds[2] + means[2]
        dh = dh * stds[3] + means[3]

    refine_center_x = nd.broadcast_add(lhs=center_x,
        rhs=nd.broadcast_mul(lhs=bbox_width, rhs=dx))
    refine_center_y = nd.broadcast_add(lhs=center_y,
        rhs=nd.broadcast_mul(lhs=bbox_height, rhs=dy))
    refined_width = nd.broadcast_mul(lhs=bbox_width, rhs=nd.exp(dw))
    refined_height = nd.broadcast_mul(lhs=bbox_height, rhs=nd.exp(dh))
    w_offset = 0.5 * (refined_width - 1.)
    h_offset = 0.5 * (refined_height - 1.)
    refined_xmin = nd.expand_dims(refine_center_x - w_offset, axis=1)
    refined_ymin = nd.expand_dims(refine_center_y - h_offset, axis=1)
    refined_xmax = nd.expand_dims(refine_center_x + w_offset, axis=1)
    refined_ymax = nd.expand_dims(refine_center_y + h_offset, axis=1)

    refined_bbox = nd.concat(refined_xmin, refined_ymin, refined_xmax, refined_ymax, dim=1)
    if im_info is not None:
        # assume im_info [[height, width, scale]] with shape (1,3)
        im_hw = nd.slice_axis(im_info, axis=1, begin=0, end=2)
        im_wh = nd.reverse(im_hw, axis=1)
        im_wh = im_wh - 1.
        im_wh = nd.tile(data=im_wh, reps=(1, 2))
        im_wh = nd.Reshape(im_wh, shape=(1, 4, 1))
        refined_bbox = nd.broadcast_minimum(lhs=refined_bbox, rhs=im_wh)
        refined_bbox = nd.broadcast_maximum(lhs=refined_bbox,
            rhs=nd.zeros_like(refined_bbox))
    # print refined_bbox.debug_str()
    return refined_bbox