def test_beta_op():
set_global_seed(1024)
_alpha, _beta = 2, 0.8
_expected_mean = _alpha / (_alpha + _beta)
_expected_std = np.sqrt(_alpha * _beta / ((_alpha + _beta)**2 *
(_alpha + _beta + 1)))
alpha = F.full([8, 9, 11, 12], value=_alpha, dtype="float32")
beta = F.full([8, 9, 11, 12], value=_beta, dtype="float32")
op = BetaRNG(seed=get_global_rng_seed())
(output, ) = apply(op, alpha, beta)
assert np.fabs(output.numpy().mean() - _expected_mean) < 1e-1
assert np.fabs(np.sqrt(output.numpy().var()) - _expected_std) < 1e-1
assert str(output.device) == str(CompNode("xpux"))
cn = CompNode("xpu2")
seed = 233333
h = new_rng_handle(cn, seed)
alpha = F.full([8, 9, 11, 12], value=_alpha, dtype="float32", device=cn)
beta = F.full([8, 9, 11, 12], value=_beta, dtype="float32", device=cn)
op = BetaRNG(seed=seed, handle=h)
(output, ) = apply(op, alpha, beta)
delete_rng_handle(h)
assert np.fabs(output.numpy().mean() - _expected_mean) < 1e-1
assert np.fabs(np.sqrt(output.numpy().var()) - _expected_std) < 1e-1
assert str(output.device) == str(cn)
def test_full():
shape = (2, 3)
values = [True, 4, 5.0]
for value in values:
np.testing.assert_allclose(
F.full(shape, value).numpy(), np.full(shape, value))
assert F.full(shape, value).dtype == tensor(value).dtype
def test_gamma_op():
set_global_seed(1024)
_shape, _scale = 2, 0.8
_expected_mean, _expected_std = _shape * _scale, np.sqrt(_shape) * _scale
shape = F.full([8, 9, 11, 12], value=_shape, dtype="float32")
scale = F.full([8, 9, 11, 12], value=_scale, dtype="float32")
op = GammaRNG(seed=get_global_rng_seed(), handle=0)
(output, ) = apply(op, shape, scale)
assert np.fabs(output.numpy().mean() - _expected_mean) < 1e-1
assert np.fabs(np.sqrt(output.numpy().var()) - _expected_std) < 1e-1
assert str(output.device) == str(CompNode("xpux"))
cn = CompNode("xpu2")
seed = 233333
h = new_rng_handle(cn, seed)
shape = F.full([8, 9, 11, 12],
value=_shape,
dtype="float32",
device="xpu2")
scale = F.full([8, 9, 11, 12],
value=_scale,
dtype="float32",
device="xpu2")
op = GammaRNG(seed=seed, handle=h)
(output, ) = apply(op, shape, scale)
delete_rng_handle(h)
assert np.fabs(output.numpy().mean() - _expected_mean) < 1e-1
assert np.fabs(np.sqrt(output.numpy().var()) - _expected_std) < 1e-1
assert str(output.device) == str(cn)
def test_device():
x = tensor([1, 2, 3], dtype="float32")
y1 = F.eye(x.shape, dtype="float32")
y2 = F.eye(x.shape, dtype="float32", device=None)
np.testing.assert_almost_equal(y1.numpy(), y2.numpy())
y3 = F.eye(x.shape, dtype="float32", device="xpux")
y4 = F.eye(x.shape, dtype="float32", device=x.device)
np.testing.assert_almost_equal(y3.numpy(), y4.numpy())
y5 = F.full((3, 2), 4, device=x.device)
y6 = F.full((3, 2), 4, device="xpux")
np.testing.assert_almost_equal(y5.numpy(), y6.numpy())
def get_padded_tensor(array: Tensor,
multiple_number: int = 32,
pad_value: float = 0) -> Tensor:
""" pad the nd-array to multiple stride of th e
Args:
array (Tensor):
the tensor with the shape of [batch, channel, height, width]
multiple_number (int):
make the height and width can be divided by multiple_number
pad_value (int): the value to be padded
Returns:
padded_array (Tensor)
"""
batch, chl, t_height, t_width = array.shape
padded_height = ((t_height + multiple_number - 1) // multiple_number *
multiple_number)
padded_width = (t_width + multiple_number -
1) // multiple_number * multiple_number
padded_array = F.full((batch, chl, padded_height, padded_width),
pad_value,
dtype=array.dtype)
ndim = array.ndim
if ndim == 4:
padded_array[:, :, :t_height, :t_width] = array
elif ndim == 3:
padded_array[:, :t_height, :t_width] = array
else:
raise Exception("Not supported tensor dim: %d" % ndim)
return padded_array
def run(zero_point, scale):
qparams = create_qparams(QuantMode.ASYMMERTIC, test_dtype, scale, zero_point)
inp_data = np.random.uniform(low=-512.0, high=512.0, size=(1, 32, 32, 32))
inp = tensor(inp_data, dtype=np.float32)
# test forward
oup = fake_quant_tensor(inp, qparams).numpy()
oup_gt = fake_quant_tensor_gt(inp, scale, zero_point, qmin, qmax).numpy()
assert np.allclose(oup, oup_gt)
assert oup.shape == oup_gt.shape
# test backward
x = tensor(inp_data, dtype=np.float32)
with Grad() as grad:
grad.wrt(x, callback=_save_to(x))
y = fake_quant_tensor(x, qparams)
grad(y, tensor(F.ones_like(x)))
x1 = tensor(inp_data, dtype=np.float32)
with Grad() as grad:
grad.wrt(x1, callback=_save_to(x1))
y1 = fake_quant_tensor_gt(x1, scale, zero_point, qmin, qmax)
grad(y1, tensor(F.ones_like(x1)))
assert np.allclose(x.grad.numpy(), x1.grad.numpy())
assert make_shape_tuple(x.grad.shape) == make_shape_tuple(x1.grad.shape)
# test nan
x = F.full((1, 32, 3, 3), np.nan)
y = fake_quant_tensor(x, qparams).numpy()
assert np.isnan(y).all()
def get_ground_truth(self, rpn_rois, im_info, gt_boxes):
if not self.training:
return rpn_rois, None, None
return_rois = []
return_labels = []
return_bbox_targets = []
# get per image proposals and gt_boxes
for bid in range(gt_boxes.shape[0]):
num_valid_boxes = im_info[bid, 4].astype("int32")
gt_boxes_per_img = gt_boxes[bid, :num_valid_boxes, :]
batch_inds = F.full((gt_boxes_per_img.shape[0], 1), bid)
gt_rois = F.concat([batch_inds, gt_boxes_per_img[:, :4]], axis=1)
batch_roi_mask = rpn_rois[:, 0] == bid
# all_rois : [batch_id, x1, y1, x2, y2]
all_rois = F.concat([rpn_rois[batch_roi_mask], gt_rois])
overlaps = layers.get_iou(all_rois[:, 1:5], gt_boxes_per_img)
max_overlaps = overlaps.max(axis=1)
gt_assignment = F.argmax(overlaps, axis=1).astype("int32")
labels = gt_boxes_per_img[gt_assignment, 4]
# ---------------- get the fg/bg labels for each roi ---------------#
fg_mask = (max_overlaps >= self.cfg.fg_threshold) & (labels >= 0)
bg_mask = ((max_overlaps >= self.cfg.bg_threshold_low)
& (max_overlaps < self.cfg.bg_threshold_high))
num_fg_rois = int(self.cfg.num_rois * self.cfg.fg_ratio)
fg_inds_mask = layers.sample_mask_from_labels(
fg_mask, num_fg_rois, 1)
num_bg_rois = int(self.cfg.num_rois - fg_inds_mask.sum())
bg_inds_mask = layers.sample_mask_from_labels(
bg_mask, num_bg_rois, 1)
labels = labels * fg_inds_mask
keep_mask = fg_inds_mask + bg_inds_mask
_, keep_inds = F.cond_take(keep_mask == 1, keep_mask)
# Add next line to avoid memory exceed
keep_inds = keep_inds[:min(self.cfg.num_rois, keep_inds.shape[0])]
labels = labels[keep_inds].astype("int32")
rois = all_rois[keep_inds]
target_boxes = gt_boxes_per_img[gt_assignment[keep_inds], :4]
bbox_targets = self.box_coder.encode(rois[:, 1:5], target_boxes)
bbox_targets = bbox_targets.reshape(-1, 4)
return_rois.append(rois)
return_labels.append(labels)
return_bbox_targets.append(bbox_targets)
return (
F.concat(return_rois, axis=0).detach(),
F.concat(return_labels, axis=0).detach(),
F.concat(return_bbox_targets, axis=0).detach(),
)
def find_top_rpn_proposals(self, rpn_cls_score_list, rpn_bbox_offset_list,
anchors_list, im_info):
prev_nms_top_n = (self.cfg.train_prev_nms_top_n
if self.training else self.cfg.test_prev_nms_top_n)
post_nms_top_n = (self.cfg.train_post_nms_top_n
if self.training else self.cfg.test_post_nms_top_n)
return_rois = []
for bid in range(im_info.shape[0]):
batch_proposal_list = []
batch_score_list = []
batch_level_list = []
for l, (rpn_cls_score, rpn_bbox_offset, anchors) in enumerate(
zip(rpn_cls_score_list, rpn_bbox_offset_list,
anchors_list)):
# get proposals and scores
offsets = rpn_bbox_offset[bid].transpose(2, 3, 0,
1).reshape(-1, 4)
proposals = self.box_coder.decode(anchors, offsets)
scores = rpn_cls_score[bid].transpose(1, 2, 0).flatten()
scores.detach()
# prev nms top n
scores, order = F.topk(scores,
descending=True,
k=prev_nms_top_n)
proposals = proposals[order, :]
batch_proposal_list.append(proposals)
batch_score_list.append(scores)
batch_level_list.append(F.full_like(scores, l))
# gather proposals, scores, level
proposals = F.concat(batch_proposal_list, axis=0)
scores = F.concat(batch_score_list, axis=0)
levels = F.concat(batch_level_list, axis=0)
proposals = layers.get_clipped_boxes(proposals, im_info[bid])
# filter invalid proposals and apply total level nms
keep_mask = layers.filter_boxes(proposals)
_, keep_inds = F.cond_take(keep_mask == 1, keep_mask)
proposals = proposals[keep_inds, :]
scores = scores[keep_inds]
levels = levels[keep_inds]
nms_keep_inds = layers.batched_nms(proposals, scores, levels,
self.cfg.rpn_nms_threshold,
post_nms_top_n)
# generate rois to rcnn head, rois shape (N, 5), info [batch_id, x1, y1, x2, y2]
rois = F.concat([proposals, scores.reshape(-1, 1)], axis=1)
rois = rois[nms_keep_inds]
batch_inds = F.full((rois.shape[0], 1), bid)
batch_rois = F.concat([batch_inds, rois[:, :4]], axis=1)
return_rois.append(batch_rois)
return_rois = F.concat(return_rois, axis=0)
return return_rois.detach()
def test_poisson_op():
set_global_seed(1024)
lam = F.full([8, 9, 11, 12], value=2, dtype="float32")
op = PoissonRNG(seed=get_global_rng_seed())
(output, ) = apply(op, lam)
assert np.fabs(output.numpy().mean() - 2.0) < 1e-1
assert np.fabs(np.sqrt(output.numpy().var()) - np.sqrt(2.0)) < 1e-1
assert str(output.device) == str(CompNode("xpux"))
cn = CompNode("xpu2")
seed = 233333
h = new_rng_handle(cn, seed)
lam = F.full([8, 9, 11, 12], value=2, dtype="float32", device=cn)
op = PoissonRNG(seed=seed, handle=h)
(output, ) = apply(op, lam)
delete_rng_handle(h)
assert np.fabs(output.numpy().mean() - 2.0) < 1e-1
assert np.fabs(np.sqrt(output.numpy().var()) - np.sqrt(2.0)) < 1e-1
assert str(output.device) == str(cn)
def test_empty_grad_in_backward():
x = mge.Parameter(F.full(100, 0.5))
y = mge.Parameter(F.ones(100))
gm = GradManager()
gm.attach([x, y])
with gm:
z = F.where(x > 0.7, x, y)
loss = z.sum()
gm.backward(loss)
assert np.all(x.grad.numpy() == 0)
assert np.all(y.grad.numpy() == 1)
def decode_outputs(self, outputs):
grids = []
strides = []
for (hsize, wsize), stride in zip(self.hw, self.strides):
xv, yv = meshgrid(F.arange(hsize), F.arange(wsize))
grid = F.stack((xv, yv), 2).reshape(1, -1, 2)
grids.append(grid)
shape = grid.shape[:2]
strides.append(F.full((*shape, 1), stride))
grids = F.concat(grids, axis=1)
strides = F.concat(strides, axis=1)
outputs[..., :2] = (outputs[..., :2] + grids) * strides
outputs[..., 2:4] = F.exp(outputs[..., 2:4]) * strides
return outputs
