Python nd.concat方法代碼示例

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def predict_batch(model, ctx, x, n_pred):
    '''
    Parameters
    ----------
    x: mx.ndarray, shape is (batch_size, 1, n_his, num_of_vertices)

    Returns
    ----------
    mx.ndarray, shape is (batch_size, 1, n_pred, num_of_vertices)
    '''
    predicts = []
    for pred_idx in range(n_pred):
        x_input = nd.concat(x, *predicts, dim=2)[:, :, - n_pred:, :]
        predicts.append(model(x_input.as_in_context(ctx))
                        .as_in_context(mx.cpu()))
    return nd.concat(*predicts, dim=2) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def extract_pairwise_multi_position_embedding_nd(position_mat, feat_dim, wave_length=1000):
    """ Extract multi-class position embedding

    Args:
        position_mat: [num_fg_classes, num_rois, num_rois, 4]
        feat_dim: dimension of embedding feature
        wave_length:

    Returns:
        embedding: [num_fg_classes, num_rois, num_rois, feat_dim]
    """
    feat_range = nd.arange(0, feat_dim / 8)
    dim_mat = nd.broadcast_power(lhs=nd.full((1,), wave_length),
                                     rhs=(8. / feat_dim) * feat_range)
    dim_mat = nd.Reshape(dim_mat, shape=(1, 1, 1, 1, -1))
    position_mat = nd.expand_dims(100.0 * position_mat, axis=4)
    div_mat = nd.broadcast_div(lhs=position_mat, rhs=dim_mat)
    sin_mat = nd.sin(data=div_mat)
    cos_mat = nd.cos(data=div_mat)
    # embedding, [num_fg_classes, num_rois, num_rois, 4, feat_dim/4]
    embedding = nd.concat(sin_mat, cos_mat, dim=4)
    embedding = nd.Reshape(embedding, shape=(0, 0, 0, feat_dim))
    return embedding 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def __next__(self):
        balanced_batch_images = []
        balanced_batch_texts = []

        for i, data_loader_iter in enumerate(self.dataloader_iter_list):
            try:
                image, text = next(data_loader_iter)
                balanced_batch_images.append(image)
                balanced_batch_texts.append(text)
            except StopIteration:
                self.dataloader_iter_list[i] = iter(self.data_loader_list[i])
                image, text = next(self.dataloader_iter_list[i])
                balanced_batch_images.append(image)
                balanced_batch_texts.append(text)
            except ValueError:
                pass

        balanced_batch_images = nd.concat(*balanced_batch_images, dim=0)
        balanced_batch_texts = nd.concat(*balanced_batch_texts, dim=0)
        return balanced_batch_images, balanced_batch_texts 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def forward(self, x):
        x_t = nd.transpose(x, axes=(0,2,1))
        conv3_out = self.conv3(x_t)
        conv5_out = self.conv5(conv3_out) + conv3_out
        conv7_out = self.conv7(conv5_out) + conv5_out 
        # conv_out = nd.concat(*[conv3_out, conv5_out, conv7_out], dim=1)
        conv_out = self.conv_drop(conv7_out)
        conv_max_pooled = self.conv_maxpool(conv_out)

        gru_out = self.gru(x)
        gru_out_t = nd.transpose(gru_out, axes=(0,2,1))
        # gru_pooled = nd.transpose(gru_out, axes=(0,2,1))
        # gru_maxpooled = self.gru_post_max(gru_out_t)
        # return gru_maxpooled
        # gru_avepooled = self.gru_post_ave(gru_out_t)
        # gru_pooled = nd.concat(*[gru_maxpooled, gru_avepooled], dim=1)

        # gru_pooled = nd.concat(*[gru_maxpooled, gru_avepooled], dim=1)
        gru_maxpooled = self.gru_maxpool(gru_out_t)
        # gru_avepooled = self.gru_maxpool(gru_out_t)
        # gru_pooled = nd.concat(*[gru_maxpooled, gru_avepooled], dim=1)

        # conv_ave_pooled = self.conv_avepool(conv_out)
        concated_feature = nd.concat(*[gru_maxpooled, conv_max_pooled], dim=1)
        return concated_feature 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def msg_edge(self, edge):
        state = nd.concat(edge.src['state'], edge.dst['state'], dim=-1)
        feature = nd.concat(edge.src['feature'], edge.dst['feature'], edge.data['dist'], dim=-1)

        # generate weight by meta-learner
        weight = self.w_mlp(feature)
        weight = nd.reshape(weight, shape=(-1, self.hidden_size * 2, self.hidden_size))

        # reshape state to [n, b * t, d] for batch_dot (currently mxnet only support batch_dot for 3D tensor)
        shape = state.shape
        state = nd.reshape(state, shape=(shape[0], -1, shape[-1]))

        alpha = nd.LeakyReLU(nd.batch_dot(state, weight))

        # reshape alpha to [n, b, t, d]
        alpha = nd.reshape(alpha, shape=shape[:-1] + (self.hidden_size,))
        return { 'alpha': alpha, 'state': edge.src['state'] } 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
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 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def test_layer_bidirectional():
    class RefBiLSTM(gluon.Block):
        def __init__(self, size, **kwargs):
            super(RefBiLSTM, self).__init__(**kwargs)
            with self.name_scope():
                self._lstm_fwd = gluon.rnn.LSTM(size, bidirectional=False, prefix='l0')
                self._lstm_bwd = gluon.rnn.LSTM(size, bidirectional=False, prefix='r0')

        def forward(self, inpt):
            fwd = self._lstm_fwd(inpt)
            bwd_inpt = nd.flip(inpt, 0)
            bwd = self._lstm_bwd(bwd_inpt)
            bwd = nd.flip(bwd, 0)
            return nd.concat(fwd, bwd, dim=2)

    size = 7
    in_size = 5
    weights = {}
    for d in ['l', 'r']:
        weights['lstm_{}0_i2h_weight'.format(d)] = mx.random.uniform(shape=(size*4, in_size))
        weights['lstm_{}0_h2h_weight'.format(d)] = mx.random.uniform(shape=(size*4, size))
        weights['lstm_{}0_i2h_bias'.format(d)] = mx.random.uniform(shape=(size*4,))
        weights['lstm_{}0_h2h_bias'.format(d)] = mx.random.uniform(shape=(size*4,))

    net = gluon.rnn.LSTM(size, bidirectional=True, prefix='lstm_')
    ref_net = RefBiLSTM(size, prefix='lstm_')
    net.initialize()
    ref_net.initialize()
    net_params = net.collect_params()
    ref_net_params = ref_net.collect_params()
    for k in weights:
        net_params[k].set_data(weights[k])
        ref_net_params[k.replace('l0', 'l0l0').replace('r0', 'r0l0')].set_data(weights[k])

    data = mx.random.uniform(shape=(3, 10, in_size))
    assert_allclose(net(data).asnumpy(), ref_net(data).asnumpy()) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def bbox_improve(bbox):
    '''bbox encoding'''
    area = (bbox[:,2] - bbox[:,0]) * (bbox[:,3] - bbox[:,1])
    return nd.concat(bbox, area.expand_dims(1)) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def forward(self, edges):
        feat = nd.concat(edges.src['pred_bbox'], edges.dst['pred_bbox'], 
                         edges.data['rel_bbox'], edges.data['pred_bbox_additional'])
        out = self.mlp(feat)
        return {'spatial': out} 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def forward(self, graph, feat):
        r"""Compute set2set pooling.

        Parameters
        ----------
        graph : DGLGraph
            The graph.
        feat : mxnet.NDArray
            The input feature with shape :math:`(N, D)` where
            :math:`N` is the number of nodes in the graph.

        Returns
        -------
        mxnet.NDArray
            The output feature with shape :math:`(B, D)`, where
            :math:`B` refers to the batch size.
        """
        with graph.local_scope():
            batch_size = graph.batch_size

            h = (nd.zeros((self.n_layers, batch_size, self.input_dim), ctx=feat.context),
                 nd.zeros((self.n_layers, batch_size, self.input_dim), ctx=feat.context))
            q_star = nd.zeros((batch_size, self.output_dim), ctx=feat.context)

            for _ in range(self.n_iters):
                q, h = self.lstm(q_star.expand_dims(axis=0), h)
                q = q.reshape((batch_size, self.input_dim))
                e = (feat * broadcast_nodes(graph, q)).sum(axis=-1, keepdims=True)
                graph.ndata['e'] = e
                alpha = softmax_nodes(graph, 'e')
                graph.ndata['r'] = feat * alpha
                readout = sum_nodes(graph, 'r')
                q_star = nd.concat(q, readout, dim=-1)

            return q_star 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def validate(val_data, val_dataset, net, ctx):
    if isinstance(ctx, mx.Context):
        ctx = [ctx]

    val_metric.reset()

    from tqdm import tqdm
    for batch in tqdm(val_data):
        data, scale, center, score, imgid = val_batch_fn(batch, ctx)

        outputs = [net(X) for X in data]
        if opt.flip_test:
            data_flip = [nd.flip(X, axis=3) for X in data]
            outputs_flip = [net(X) for X in data_flip]
            outputs_flipback = [flip_heatmap(o, val_dataset.joint_pairs, shift=True) for o in outputs_flip]
            outputs = [(o + o_flip)/2 for o, o_flip in zip(outputs, outputs_flipback)]

        if opt.dsnt:
            outputs = [net_dsnt(X)[0] for X in outputs]

        if len(outputs) > 1:
            outputs_stack = nd.concat(*[o.as_in_context(mx.cpu()) for o in outputs], dim=0)
        else:
            outputs_stack = outputs[0].as_in_context(mx.cpu())

        if opt.dsnt:
            preds = (outputs_stack - 0.5) * scale.expand_dims(axis=1) + center.expand_dims(axis=1)
            maxvals = nd.ones(preds.shape[0:2]+(1, ))
        else:
            preds, maxvals = get_final_preds(outputs_stack, center.asnumpy(), scale.asnumpy())
        val_metric.update(preds, maxvals, score, imgid)

    metric_name, metric_score = val_metric.get()
    print("Inference Completed! %s = %.4f" % (metric_name, metric_score))
    return 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def validate(val_data, val_dataset, net, ctx):
    if isinstance(ctx, mx.Context):
        ctx = [ctx]

    val_metric.reset()

    from tqdm import tqdm
    for batch in tqdm(val_data):
        # data, scale, center, score, imgid = val_batch_fn(batch, ctx)
        data, scale_box, score, imgid = val_batch_fn(batch, ctx)

        outputs = [net(X) for X in data]
        if opt.flip_test:
            data_flip = [nd.flip(X, axis=3) for X in data]
            outputs_flip = [net(X) for X in data_flip]
            outputs_flipback = [flip_heatmap(o, val_dataset.joint_pairs, shift=True) for o in outputs_flip]
            outputs = [(o + o_flip)/2 for o, o_flip in zip(outputs, outputs_flipback)]

        if len(outputs) > 1:
            outputs_stack = nd.concat(*[o.as_in_context(mx.cpu()) for o in outputs], dim=0)
        else:
            outputs_stack = outputs[0].as_in_context(mx.cpu())

        # preds, maxvals = get_final_preds(outputs_stack, center.asnumpy(), scale.asnumpy())
        preds, maxvals = heatmap_to_coord_alpha_pose(outputs_stack, scale_box)
        # print(preds, maxvals, scale_box)
        # print(preds, maxvals)
        # raise
        val_metric.update(preds, maxvals, score, imgid)

    res = val_metric.get()
    return 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def validate(val_data, val_dataset, net, ctx, opt):
    if isinstance(ctx, mx.Context):
        ctx = [ctx]

    val_metric = COCOKeyPointsMetric(val_dataset, 'coco_keypoints',
                                     in_vis_thresh=0)

    for batch in tqdm(val_data, dynamic_ncols=True):
        # data, scale, center, score, imgid = val_batch_fn(batch, ctx)
        data, scale_box, score, imgid = val_batch_fn(batch, ctx)

        outputs = [net(X) for X in data]
        if opt.flip_test:
            data_flip = [nd.flip(X, axis=3) for X in data]
            outputs_flip = [net(X) for X in data_flip]
            outputs_flipback = [flip_heatmap(o, val_dataset.joint_pairs, shift=True) for o in outputs_flip]
            outputs = [(o + o_flip) / 2 for o, o_flip in zip(outputs, outputs_flipback)]

        if len(outputs) > 1:
            outputs_stack = nd.concat(*[o.as_in_context(mx.cpu()) for o in outputs], dim=0)
        else:
            outputs_stack = outputs[0].as_in_context(mx.cpu())

        # preds, maxvals = get_final_preds(outputs_stack, center.asnumpy(), scale.asnumpy())
        preds, maxvals = heatmap_to_coord_alpha_pose(outputs_stack, scale_box)
        val_metric.update(preds, maxvals, score, imgid)

    nullwriter = NullWriter()
    oldstdout = sys.stdout
    sys.stdout = nullwriter
    try:
        res = val_metric.get()
    finally:
        sys.stdout = oldstdout
    return res 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def _slice(self, x, num_anchors, num_offsets):
        """since some stages won't see partial anchors, so we have to slice the correct targets"""
        # x with shape (B, N, A, 1 or 2)
        anchors = [0] + num_anchors.tolist()
        offsets = [0] + num_offsets.tolist()
        ret = []
        for i in range(len(num_anchors)):
            y = x[:, offsets[i]:offsets[i+1], anchors[i]:anchors[i+1], :]
            ret.append(y.reshape((0, -3, -1)))
        return nd.concat(*ret, dim=1) 

# 需要導入模塊: from mxnet import nd [as 別名]
# 或者: from mxnet.nd import concat [as 別名]
def get_final_preds(batch_heatmaps, center, scale):
    coords, maxvals = get_max_pred(batch_heatmaps)

    heatmap_height = batch_heatmaps.shape[2]
    heatmap_width = batch_heatmaps.shape[3]

    # post-processing
    for n in range(coords.shape[0]):
        for p in range(coords.shape[1]):
            hm = batch_heatmaps[n][p]
            px = int(nd.floor(coords[n][p][0] + 0.5).asscalar())
            py = int(nd.floor(coords[n][p][1] + 0.5).asscalar())
            if 1 < px < heatmap_width-1 and 1 < py < heatmap_height-1:
                diff = nd.concat(hm[py][px+1] - hm[py][px-1],
                                 hm[py+1][px] - hm[py-1][px],
                                 dim=0)
                coords[n][p] += nd.sign(diff) * .25

    preds = nd.zeros_like(coords)

    # Transform back
    for i in range(coords.shape[0]):
        preds[i] = transform_preds(coords[i], center[i], scale[i],
                                   [heatmap_width, heatmap_height])

    return preds, maxvals