Python mxnet.nd.tile() Examples

def get_max_pred(batch_heatmaps):
    batch_size = batch_heatmaps.shape[0]
    num_joints = batch_heatmaps.shape[1]
    width = batch_heatmaps.shape[3]
    heatmaps_reshaped = batch_heatmaps.reshape((batch_size, num_joints, -1))
    idx = nd.argmax(heatmaps_reshaped, 2)
    maxvals = nd.max(heatmaps_reshaped, 2)

    maxvals = maxvals.reshape((batch_size, num_joints, 1))
    idx = idx.reshape((batch_size, num_joints, 1))

    preds = nd.tile(idx, (1, 1, 2)).astype(np.float32)

    preds[:, :, 0] = (preds[:, :, 0]) % width
    preds[:, :, 1] = nd.floor((preds[:, :, 1]) / width)

    pred_mask = nd.tile(nd.greater(maxvals, 0.0), (1, 1, 2))
    pred_mask = pred_mask.astype(np.float32)

    preds *= pred_mask
    return preds, maxvals 

def get_max_pred(batch_heatmaps):
    batch_size = batch_heatmaps.shape[0]
    num_joints = batch_heatmaps.shape[1]
    width = batch_heatmaps.shape[3]
    heatmaps_reshaped = batch_heatmaps.reshape((batch_size, num_joints, -1))
    idx = nd.argmax(heatmaps_reshaped, 2)
    maxvals = nd.max(heatmaps_reshaped, 2)

    maxvals = maxvals.reshape((batch_size, num_joints, 1))
    idx = idx.reshape((batch_size, num_joints, 1))

    preds = nd.tile(idx, (1, 1, 2)).astype(np.float32)

    preds[:, :, 0] = (preds[:, :, 0]) % width
    preds[:, :, 1] = nd.floor((preds[:, :, 1]) / width)

    pred_mask = nd.tile(nd.greater(maxvals, 0.0), (1, 1, 2))
    pred_mask = pred_mask.astype(np.float32)

    preds *= pred_mask
    return preds, maxvals 

def hybrid_forward(self, F, X, y=None):
        # import pdb; pdb.set_trace()
        X = self.net[0](X) # Conv1
        X = self.net[1](X) # Primary Capsule
        X = self.net[2](X) # Digital Capsule
        # import pdb ; pdb.set_trace()
        X = X.reshape((X.shape[0],X.shape[2], X.shape[4]))
        # get length of vector for margin loss calculation
        X_l2norm = nd.sqrt((X**2).sum(axis=-1))
        # import pdb ; pdb.set_trace()
        prob = nd.softmax(X_l2norm, axis=-1)

        if y is not None:
            max_len_indices = y
        else:
            
            max_len_indices = nd.argmax(prob,axis=-1)


        y_tile = nd.tile(y.expand_dims(axis=1), reps=(1, X.shape[-1]))
        batch_activated_capsules = nd.pick(X, y_tile, axis=1, keepdims=True)

        reconstrcutions = self.net[3](batch_activated_capsules)

        return  prob, X_l2norm, reconstrcutions 

def test_radial_basis_function_kernel(
    x1, x2, amplitude, length_scale, exact
) -> None:
    tol = 1e-5
    batch_size = amplitude.shape[0]
    history_length_1 = x1.shape[0]
    history_length_2 = x2.shape[0]
    num_features = x1.shape[1]
    if batch_size > 1:
        x1 = nd.tile(x1, reps=(batch_size, 1, 1))
        x2 = nd.tile(x2, reps=(batch_size, 1, 1))
        for i in range(1, batch_size):
            x1[i, :, :] = (i + 1) * x1[i, :, :]
            x2[i, :, :] = (i - 3) * x2[i, :, :]
    else:
        x1 = x1.reshape(batch_size, history_length_1, num_features)
        x2 = x2.reshape(batch_size, history_length_2, num_features)
    amplitude = amplitude.reshape(batch_size, 1, 1)
    length_scale = length_scale.reshape(batch_size, 1, 1)
    rbf = RBFKernel(amplitude, length_scale)

    exact = amplitude * nd.exp(-0.5 * exact / length_scale ** 2)

    res = rbf.kernel_matrix(x1, x2)
    assert nd.norm(exact - res) < tol 

def transform(data, target_wd, target_ht, is_train, box):
    """Crop and normnalize an image nd array."""
    if box is not None:
        x, y, w, h = box
        data = data[y:min(y+h, data.shape[0]), x:min(x+w, data.shape[1])]

    # Resize to target_wd * target_ht.
    data = mx.image.imresize(data, target_wd, target_ht)

    # Normalize in the same way as the pre-trained model.
    data = data.astype(np.float32) / 255.0
    data = (data - mx.nd.array([0.485, 0.456, 0.406])) / mx.nd.array([0.229, 0.224, 0.225])

    if is_train:
        if random.random() < 0.5:
            data = nd.flip(data, axis=1)
        data, _ = mx.image.random_crop(data, (224, 224))
    else:
        data, _ = mx.image.center_crop(data, (224, 224))

    # Transpose from (target_wd, target_ht, 3)
    # to (3, target_wd, target_ht).
    data = nd.transpose(data, (2, 0, 1))

    # If image is greyscale, repeat 3 times to get RGB image.
    if data.shape[0] == 1:
        data = nd.tile(data, (3, 1, 1))
    return data.reshape((1,) + data.shape) 

def transform(data, target_wd, target_ht, is_train, box):
    """Crop and normnalize an image nd array."""
    if box is not None:
        x, y, w, h = box
        data = data[y:min(y+h, data.shape[0]), x:min(x+w, data.shape[1])]

    # Resize to target_wd * target_ht.
    data = mx.image.imresize(data, target_wd, target_ht)

    # Normalize in the same way as the pre-trained model.
    data = data.astype(np.float32) / 255.0
    data = (data - mx.nd.array([0.485, 0.456, 0.406])) / mx.nd.array([0.229, 0.224, 0.225])

    if is_train:
        if random.random() < 0.5:
            data = nd.flip(data, axis=1)
        data, _ = mx.image.random_crop(data, (224, 224))
    else:
        data, _ = mx.image.center_crop(data, (224, 224))

    # Transpose from (target_wd, target_ht, 3)
    # to (3, target_wd, target_ht).
    data = nd.transpose(data, (2, 0, 1))

    # If image is greyscale, repeat 3 times to get RGB image.
    if data.shape[0] == 1:
        data = nd.tile(data, (3, 1, 1))
    return data.reshape((1,) + data.shape) 

def test_periodic_kernel(x1, x2, amplitude, length_scale, exact) -> None:
    tol = 1e-5
    batch_size = amplitude.shape[0]
    history_length_1 = x1.shape[0]
    history_length_2 = x2.shape[0]
    num_features = x1.shape[1]
    if batch_size > 1:
        x1 = nd.tile(x1, reps=(batch_size, 1, 1))
        x2 = nd.tile(x2, reps=(batch_size, 1, 1))
        for i in range(1, batch_size):
            x1[i, :, :] = (i + 1) * x1[i, :, :]
            x2[i, :, :] = (i - 3) * x2[i, :, :]
    else:
        x1 = x1.reshape(batch_size, history_length_1, num_features)
        x2 = x2.reshape(batch_size, history_length_2, num_features)
    amplitude = amplitude.reshape(batch_size, 1, 1)
    length_scale = length_scale.reshape(batch_size, 1, 1)
    frequency = 1 / 24 * nd.ones_like(length_scale)
    periodic = PeriodicKernel(amplitude, length_scale, frequency)

    exact = amplitude * nd.exp(
        -2
        * nd.sin(frequency * math.pi * nd.sqrt(exact)) ** 2
        / length_scale ** 2
    )

    res = periodic.kernel_matrix(x1, x2)
    assert nd.norm(exact - res) < tol


# This test is based off of a simple single batch with different history lengths
# from gpytorch, where the exact is computed inside the test rather than hard-coded 

def transform(data, target_wd, target_ht, is_train, box):
    """Crop and normnalize an image nd array."""
    if box is not None:
        x, y, w, h = box
        data = data[y:min(y+h, data.shape[0]), x:min(x+w, data.shape[1])]

    # Resize to target_wd * target_ht.
    data = mx.image.imresize(data, target_wd, target_ht)

    # Normalize in the same way as the pre-trained model.
    data = data.astype(np.float32) / 255.0
    data = (data - mx.nd.array([0.485, 0.456, 0.406])) / mx.nd.array([0.229, 0.224, 0.225])

    if is_train:
        if random.random() < 0.5:
            data = nd.flip(data, axis=1)
        data, _ = mx.image.random_crop(data, (224, 224))
    else:
        data, _ = mx.image.center_crop(data, (224, 224))

    # Transpose from (target_wd, target_ht, 3)
    # to (3, target_wd, target_ht).
    data = nd.transpose(data, (2, 0, 1))

    # If image is greyscale, repeat 3 times to get RGB image.
    if data.shape[0] == 1:
        data = nd.tile(data, (3, 1, 1))
    return data.reshape((1,) + data.shape) 

def random_expand(src, max_ratio=4, fill=0, keep_ratio=True):
    """Random expand original image with borders, this is identical to placing
    the original image on a larger canvas.

    Parameters
    ----------
    src : mxnet.nd.NDArray
        The original image with HWC format.
    max_ratio : int or float
        Maximum ratio of the output image on both direction(vertical and horizontal)
    fill : int or float or array-like
        The value(s) for padded borders. If `fill` is numerical type, RGB channels
        will be padded with single value. Otherwise `fill` must have same length
        as image channels, which resulted in padding with per-channel values.
    keep_ratio : bool
        If `True`, will keep output image the same aspect ratio as input.

    Returns
    -------
    mxnet.nd.NDArray
        Augmented image.
    tuple
        Tuple of (offset_x, offset_y, new_width, new_height)

    """
    if max_ratio <= 1:
        return src, (0, 0, src.shape[1], src.shape[0])

    h, w, c = src.shape
    ratio_x = random.uniform(1, max_ratio)
    if keep_ratio:
        ratio_y = ratio_x
    else:
        ratio_y = random.uniform(1, max_ratio)

    oh, ow = int(h * ratio_y), int(w * ratio_x)
    off_y = random.randint(0, oh - h)
    off_x = random.randint(0, ow - w)

    # make canvas
    if isinstance(fill, numeric_types):
        dst = nd.full(shape=(oh, ow, c), val=fill, dtype=src.dtype)
    else:
        fill = nd.array(fill, dtype=src.dtype, ctx=src.context)
        if not c == fill.size:
            raise ValueError("Channel and fill size mismatch, {} vs {}".format(c, fill.size))
        dst = nd.tile(fill.reshape((1, c)), reps=(oh * ow, 1)).reshape((oh, ow, c))

    dst[off_y:off_y+h, off_x:off_x+w, :] = src
    return dst, (off_x, off_y, ow, oh) 

def random_expand(src, max_ratio=4, fill=0, keep_ratio=True):
    """Random expand original image with borders, this is identical to placing
    the original image on a larger canvas.

    Parameters
    ----------
    src : mxnet.nd.NDArray
        The original image with HWC format.
    max_ratio : int or float
        Maximum ratio of the output image on both direction(vertical and horizontal)
    fill : int or float or array-like
        The value(s) for padded borders. If `fill` is numerical type, RGB channels
        will be padded with single value. Otherwise `fill` must have same length
        as image channels, which resulted in padding with per-channel values.
    keep_ratio : bool
        If `True`, will keep output image the same aspect ratio as input.

    Returns
    -------
    mxnet.nd.NDArray
        Augmented image.
    tuple
        Tuple of (offset_x, offset_y, new_width, new_height)

    """
    if max_ratio <= 1:
        return src, (0, 0, src.shape[1], src.shape[0])

    h, w, c = src.shape
    ratio_x = random.uniform(1, max_ratio)
    if keep_ratio:
        ratio_y = ratio_x
    else:
        ratio_y = random.uniform(1, max_ratio)

    oh, ow = int(h * ratio_y), int(w * ratio_x)
    off_y = random.randint(0, oh - h)
    off_x = random.randint(0, ow - w)

    # make canvas
    if isinstance(fill, numeric_types):
        dst = nd.full(shape=(oh, ow, c), val=fill, dtype=src.dtype)
    else:
        fill = nd.array(fill, dtype=src.dtype, ctx=src.context)
        if not c == fill.size:
            raise ValueError("Channel and fill size mismatch, {} vs {}".format(c, fill.size))
        dst = nd.tile(fill.reshape((1, c)), reps=(oh * ow, 1)).reshape((oh, ow, c))

    dst[off_y:off_y+h, off_x:off_x+w, :] = src
    return dst, (off_x, off_y, ow, oh) 

def random_expand(src, max_ratio=4, fill=0, keep_ratio=True):
    """Random expand original image with borders, this is identical to placing
    the original image on a larger canvas.

    Parameters
    ----------
    src : mxnet.nd.NDArray
        The original image with HWC format.
    max_ratio : int or float
        Maximum ratio of the output image on both direction(vertical and horizontal)
    fill : int or float or array-like
        The value(s) for padded borders. If `fill` is numerical type, RGB channels
        will be padded with single value. Otherwise `fill` must have same length
        as image channels, which resulted in padding with per-channel values.
    keep_ratio : bool
        If `True`, will keep output image the same aspect ratio as input.

    Returns
    -------
    mxnet.nd.NDArray
        Augmented image.
    tuple
        Tuple of (offset_x, offset_y, new_width, new_height)

    """
    if max_ratio <= 1:
        return src, (0, 0, src.shape[1], src.shape[0])

    h, w, c = src.shape
    ratio_x = random.uniform(1, max_ratio)
    if keep_ratio:
        ratio_y = ratio_x
    else:
        ratio_y = random.uniform(1, max_ratio)

    oh, ow = int(h * ratio_y), int(w * ratio_x)
    off_y = random.randint(0, oh - h)
    off_x = random.randint(0, ow - w)

    # make canvas
    if isinstance(fill, numeric_types):
        dst = nd.full(shape=(oh, ow, c), val=fill, dtype=src.dtype)
    else:
        fill = nd.array(fill, dtype=src.dtype, ctx=src.context)
        if not c == fill.size:
            raise ValueError("Channel and fill size mismatch, {} vs {}".format(c, fill.size))
        dst = nd.tile(fill.reshape((1, c)), reps=(oh * ow, 1)).reshape((oh, ow, c))

    dst[off_y:off_y+h, off_x:off_x+w, :] = src
    return dst, (off_x, off_y, ow, oh) 

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 

def random_expand(src, max_ratio=4, fill=0, keep_ratio=True):
    """Random expand original image with borders, this is identical to placing
    the original image on a larger canvas.

    Parameters
    ----------
    src : mxnet.nd.NDArray
        The original image with HWC format.
    max_ratio : int or float
        Maximum ratio of the output image on both direction(vertical and horizontal)
    fill : int or float or array-like
        The value(s) for padded borders. If `fill` is numerical type, RGB channels
        will be padded with single value. Otherwise `fill` must have same length
        as image channels, which resulted in padding with per-channel values.
    keep_ratio : bool
        If `True`, will keep output image the same aspect ratio as input.

    Returns
    -------
    mxnet.nd.NDArray
        Augmented image.
    tuple
        Tuple of (offset_x, offset_y, new_width, new_height)

    """
    if max_ratio <= 1:
        return src, (0, 0, src.shape[1], src.shape[0])

    h, w, c = src.shape
    ratio_x = random.uniform(1, max_ratio)
    if keep_ratio:
        ratio_y = ratio_x
    else:
        ratio_y = random.uniform(1, max_ratio)

    oh, ow = int(h * ratio_y), int(w * ratio_x)
    off_y = random.randint(0, oh - h)
    off_x = random.randint(0, ow - w)

    # make canvas
    if isinstance(fill, numeric_types):
        dst = nd.full(shape=(oh, ow, c), val=fill, dtype=src.dtype)
    else:
        fill = nd.array(fill, dtype=src.dtype, ctx=src.context)
        if not c == fill.size:
            raise ValueError("Channel and fill size mismatch, {} vs {}".format(c, fill.size))
        dst = nd.tile(fill.reshape((1, c)), reps=(oh * ow, 1)).reshape((oh, ow, c))

    dst[off_y:off_y+h, off_x:off_x+w, :] = src
    return dst, (off_x, off_y, ow, oh)