p27049618
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

All the descriptions of batch normal under the mindspore folder have been uniformly updated to 'Batch Normalization'

This commit is contained in:
dinglinhe 2021-04-19 11:19:53 +08:00
parent 4631facee9
commit 865cf68243
11 changed files with 40 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
mindspore/ccsrc/backend/kernel_compiler/cpu/mkldnn/batch_norm_gard_cpu_kernel.cc
View File

@ -51,12 +51,12 @@ void BatchNormGradCPUKernel::InitKernel(const CNodePtr &kernel_node) {
auto prop_kind = dnnl::prop_kind::forward_training;
auto normalization_flags = dnnl::normalization_flags::use_scale_shift;
// fused batch normalization forward description
// fused Batch Normalization forward description
dnnl::batch_normalization_forward::desc desc =
dnnl::batch_normalization_forward::desc(prop_kind, x_desc, epsilon, normalization_flags);
auto forward_prim_desc = dnnl::batch_normalization_forward::primitive_desc(desc, MKLKernelEngine::Get().engine());
// fused batch normalization backward description
// fused Batch Normalization backward description
dnnl::batch_normalization_backward::desc backward_desc =
dnnl::batch_normalization_backward::desc(dnnl::prop_kind::backward, x_desc, x_desc, epsilon, normalization_flags);
auto backward_prim_desc = dnnl::batch_normalization_backward::primitive_desc(

2
mindspore/ccsrc/transform/express_ir/onnx_exporter.cc
View File

@ -33,7 +33,7 @@ enum OpMergeMode {
OP_MERGE_IGNORE = 1, // indicate an input op merged into other op in compute node list
OP_MERGE_CONV = 2, // indicate `MindSpore Conv + BiasAdd` --> `ONNX Conv`
OP_MERGE_GEMM = 3, // indicate `MindSpore MatMul + BiasAdd` --> `ONNX Gemm`
OP_MERGE_BATCH_NORM = 4, // indicate `MindSpore BatchNorm(x)[0]` --> `ONNX BatchNormalization`
OP_MERGE_BATCH_NORM = 4, // indicate `MindSpore BatchNorm(x)[0]` --> `ONNX Batch Normalization`
OP_MERGE_MAXPOOL_WITH_ARGMAX = 5, // indicate `MindSpore MaxPoolWithArgmax(x)[0]` --> `ONNX MaxPool`
};

4
mindspore/compression/quant/qat.py
View File

@ -339,7 +339,7 @@ class QuantizationAwareTraining(Quantizer):
quant_config=self.quant_config,
quant_dtype=self.weight_dtype,
fake=True)
# change original network BatchNormal OP parameters to quant network
# change original network Batch Normalization OP parameters to quant network
conv_inner.gamma = subcell.batchnorm.gamma
conv_inner.beta = subcell.batchnorm.beta
conv_inner.moving_mean = subcell.batchnorm.moving_mean
@ -363,7 +363,7 @@ class QuantizationAwareTraining(Quantizer):
bias_init=conv_inner.bias_init,
quant_config=self.quant_config,
quant_dtype=self.weight_dtype)
# change original network BatchNormal OP parameters to quant network
# change original network Batch Normalization OP parameters to quant network
conv_inner.batchnorm.gamma = subcell.batchnorm.gamma
conv_inner.batchnorm.beta = subcell.batchnorm.beta
conv_inner.batchnorm.moving_mean = subcell.batchnorm.moving_mean

2
mindspore/nn/acc/less_batch_normalization.py
View File

@ -12,7 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""less batch normalization"""
"""less Batch Normalization"""
import numpy as np
from mindspore import nn
from mindspore.ops import operations as P

16
mindspore/nn/layer/normalization.py
View File

@ -233,7 +233,7 @@ def _shape_check_bn(in_shape, in_dims):
@constexpr
def _shape_infer(x_shape, num_feature):
"""global batch normalization shape and axes infer"""
"""global Batch Normalization shape and axes infer"""
if len(x_shape) == 4:
axes = (0, 2, 3)
re_shape = (1, num_feature, 1, 1)
@ -245,7 +245,7 @@ def _shape_infer(x_shape, num_feature):
class BatchNorm1d(_BatchNorm):
r"""
Batch normalization layer over a 2D input.
Batch Normalization layer over a 2D input.
Batch Normalization is widely used in convolutional networks. This layer
applies Batch Normalization over a 2D input (a mini-batch of 1D inputs) to
@ -334,7 +334,7 @@ class BatchNorm1d(_BatchNorm):
class BatchNorm2d(_BatchNorm):
r"""
Batch normalization layer over a 4D input.
Batch Normalization layer over a 4D input.
Batch Normalization is widely used in convolutional networks. This layer
applies Batch Normalization over a 4D input (a mini-batch of 2D inputs with
@ -441,7 +441,7 @@ def _check_3d_shape(input_shape):
class BatchNorm3d(Cell):
r"""
Batch normalization layer over a 5D input.
Batch Normalization layer over a 5D input.
Batch Normalization is widely used in convolutional networks. This layer
applies Batch Normalization over a 5D input (a mini-batch of 3D inputs with
@ -540,7 +540,7 @@ class GlobalBatchNorm(_BatchNorm):
r"""
Global normalization layer over a N-dimension input.
Global Normalization is cross device synchronized batch normalization. The implementation of Batch Normalization
Global Normalization is cross device synchronized Batch Normalization. The implementation of Batch Normalization
only normalizes the data within each device. Global normalization will normalize the input within the group.
It has been described in the paper `Batch Normalization: Accelerating Deep Network Training by
Reducing Internal Covariate Shift <https://arxiv.org/abs/1502.03167>`_. It rescales and recenters the
@ -653,10 +653,10 @@ class GlobalBatchNorm(_BatchNorm):
class SyncBatchNorm(_BatchNorm):
r"""
Sync Batch normalization layer over a N-dimension input.
Sync Batch Normalization layer over a N-dimension input.
Sync Batch Normalization is cross device synchronized batch normalization. The implementation of Batch
Normalization only normalizes the data within each device. Sync Batch normalization will normalize the input
Sync Batch Normalization is cross device synchronized Batch Normalization. The implementation of Batch
Normalization only normalizes the data within each device. Sync Batch Normalization will normalize the input
within the group. It has been described in the paper `Batch Normalization: Accelerating Deep Network Training by
Reducing Internal Covariate Shift <https://arxiv.org/abs/1502.03167>`_. It rescales and recenters the
feature using a mini-batch of data and the learned parameters which can be described in the following formula.

22
mindspore/nn/layer/quant.py
View File

@ -47,7 +47,7 @@ __all__ = [
class BatchNormFoldCell(Cell):
"""
Batch normalization folded.
Batch Normalization folded.
Args:
momentum (float): Momentum value must be [0, 1]. Default: 0.9.
@ -402,7 +402,8 @@ quant_config_default = QuantConfig(weight=FakeQuantWithMinMaxObserver, activatio
class Conv2dBnFoldQuantOneConv(Cell):
r"""
2D convolution which use the convolution layer statistics once to calculate BatchNormal operation folded construct.
2D convolution which use the convolution layer statistics once to calculate Batch Normalization
operation folded construct.
This part is a more detailed overview of Conv2d operation. For more detials about Quantilization,
please refer to :class:`mindspore.nn.FakeQuantWithMinMaxObserver`.
@ -414,8 +415,8 @@ class Conv2dBnFoldQuantOneConv(Cell):
stride (int): Specifies stride for all spatial dimensions with the same value.
pad_mode (str): Specifies padding mode. The optional values are "same", "valid", "pad". Default: "same".
padding (int): Implicit paddings on both sides of the input. Default: 0.
eps (float): Parameters for BatchNormal. Default: 1e-5.
momentum (float): Parameters for BatchNormal op. Default: 0.997.
eps (float): Parameters for Batch Normalization. Default: 1e-5.
momentum (float): Parameters for Batch Normalization op. Default: 0.997.
dilation (int): Specifies the dilation rate to use for dilated convolution. Default: 1.
group (int): Splits filter into groups, `in_ channels` and `out_channels` must be
divisible by the number of groups. Default: 1.
@ -600,7 +601,7 @@ class Conv2dBnFoldQuantOneConv(Cell):
class Conv2dBnFoldQuant(Cell):
r"""
2D convolution with BatchNormal operation folded construct.
2D convolution with Batch Normalization operation folded construct.
This part is a more detailed overview of Conv2d operation. For more detials about Quantilization,
please refer to :class:`mindspore.nn.FakeQuantWithMinMaxObserver`.
@ -612,8 +613,8 @@ class Conv2dBnFoldQuant(Cell):
stride (int): Specifies stride for all spatial dimensions with the same value.
pad_mode (str): Specifies padding mode. The optional values are "same", "valid", "pad". Default: "same".
padding (int): Implicit paddings on both sides of the input. Default: 0.
eps (float): Parameters for BatchNormal. Default: 1e-5.
momentum (float): Parameters for BatchNormal op. Default: 0.997.
eps (float): Parameters for Batch Normalization. Default: 1e-5.
momentum (float): Parameters for Batch Normalization op. Default: 0.997.
dilation (int): Specifies the dilation rate to use for dilated convolution. Default: 1.
group (int): Splits filter into groups, `in_ channels` and `out_channels` must be
divisible by the number of groups. Default: 1.
@ -635,7 +636,8 @@ class Conv2dBnFoldQuant(Cell):
generated by compression.quant.create_quant_config method.
Default: both set to default FakeQuantWithMinMaxObserver.
quant_dtype (QuantDtype): Specifies the FakeQuant datatype. Default: QuantDtype.INT8.
freeze_bn (int): The quantization freeze BatchNormal op is according to the global step. Default: 100000.
freeze_bn (int): The quantization freeze Batch Normalization op is according to the global step.
Default: 100000.
Inputs:
- **input** (Tensor) - Tensor of shape :math:`(N, C_{in}, H_{in}, W_{in})`.
@ -811,8 +813,8 @@ class Conv2dBnWithoutFoldQuant(Cell):
group (int): Splits filter into groups, `in_ channels` and `out_channels` must be
divisible by the number of groups. Default: 1.
has_bias (bool): Specifies whether the layer uses a bias vector. Default: False.
eps (float): Parameters for BatchNormal. Default: 1e-5.
momentum (float): Parameters for BatchNormal op. Default: 0.997.
eps (float): Parameters for Batch Normalization. Default: 1e-5.
momentum (float): Parameters for Batch Normalization op. Default: 0.997.
weight_init (Union[Tensor, str, Initializer, numbers.Number]): Initializer for the convolution kernel.
Default: 'normal'.
bias_init (Union[Tensor, str, Initializer, numbers.Number]): Initializer for the bias vector. Default: 'zeros'.

2
mindspore/nn/layer/timedistributed.py
View File

@ -65,7 +65,7 @@ class TimeDistributed(Cell):
There are two cases in the implementation.
When reshape_with_axis provided, the reshape method will be chosen, which is more efficient;
otherwise, the method of dividing the inputs along time axis will be used, which is more general.
For example, reshape_with_axis could not be provided when deal with batch normal.
For example, reshape_with_axis could not be provided when deal with Batch Normalization.
Args:
layer(Union[Cell, Primitive]): The Cell or Primitive which will be wrapped.

2
mindspore/ops/operations/_inner_ops.py
View File

@ -739,7 +739,7 @@ class SyncBatchNorm(PrimitiveWithInfer):
r"""
Sync Batch Normalization for input data and updated parameters.
Sync Batch Normalization is cross device synchronized batch normalization. Batch Normalization is
Sync Batch Normalization is cross device synchronized Batch Normalization. Batch Normalization is
widely used in convolutional neural networks. This operation applies Batch Normalization over input
to avoid internal covariate shift as described in the paper `Batch Normalization: Accelerating
Deep Network Training by Reducing Internal Covariate Shift <https://arxiv.org/abs/1502.03167>`_.

6
mindspore/ops/operations/_quant_ops.py
View File

@ -411,8 +411,8 @@ class FakeQuantPerLayer(PrimitiveWithInfer):
ema (bool): Uses EMA algorithm update value min and max. Default: False.
ema_decay (int) : EMA algorithm decay parameter. Default: 0.999.
quant_delay (int): Quantilization delay parameter. Before delay step in training time not update
simulate quantization aware funcion. After delay step in training time begin simulate the aware
quantize funcion. Default: 0.
simulate quantization aware function. After delay step in training time begin simulate the aware
quantize function. Default: 0.
symmetric (bool): Whether the quantization algorithm is symmetric or not. Default: False.
narrow_range (bool): Whether the quantization algorithm uses narrow range or not. Default: False.
training (bool): Training the network or not. Default: True.
@ -687,7 +687,7 @@ class FakeQuantPerChannelGrad(PrimitiveWithInfer):
class BatchNormFold(PrimitiveWithInfer):
"""
Batch normalization folded.
Batch Normalization folded.
Args:
momentum (float): Momentum value must be [0, 1]. Default: 0.9.

4
model_zoo/official/cv/vgg16/src/vgg.py
View File

@ -129,7 +129,7 @@ cfg = {
def vgg16(num_classes=1000, args=None, phase="train", **kwargs):
"""
Get Vgg16 neural network with batch normalization.
Get Vgg16 neural network with Batch Normalization.
Args:
num_classes (int): Class numbers. Default: 1000.
@ -137,7 +137,7 @@ def vgg16(num_classes=1000, args=None, phase="train", **kwargs):
phase(str): train or test mode.
Returns:
Cell, cell instance of Vgg16 neural network with batch normalization.
Cell, cell instance of Vgg16 neural network with Batch Normalization.
Examples:
>>> vgg16(num_classes=1000, args=args, **kwargs)

14
tests/vm_impl/vm_me.py
View File

@ -57,7 +57,7 @@ def avg_pool_grad(dout, origin_shape, pool_h, pool_w, stride):
stride (int): The stride of the sliding window.
Returns:
numpy.ndarray, grad of avgerage pooling.
numpy.ndarray, grad of average pooling.
"""
# pylint: disable=unused-argument
_, _, height, width = dout.shape
@ -70,7 +70,7 @@ def avg_pool_grad(dout, origin_shape, pool_h, pool_w, stride):
def _batch_norm(x, scale, shift, running_mean=None, running_var=None,
eps=1e-05, momentum=0.1, is_training=True):
"""Batch normalization over an array."""
"""Batch Normalization over an array."""
_, c_h_w = x.shape
# Handle running_mean and running_var are not None
# if running_mean is None:
@ -106,7 +106,7 @@ def _batch_norm(x, scale, shift, running_mean=None, running_var=None,
def batch_norm(x, scale=1, shift=0, mean=None, variance=None,
eps=1e-05, momentum=0.1, is_training=True):
"""Batch normalization over an array."""
"""Batch Normalization over an array."""
input_shape = x.shape
if x.ndim != 2:
batch_num = x.shape[0]
@ -120,7 +120,7 @@ def batch_norm(x, scale=1, shift=0, mean=None, variance=None,
def _batch_norm_grad(dout, x, scale, save_mean, save_inv_variance, \
eps=1e-05, momentum=0.1, is_training=True):
"""Batch normalization over an array."""
"""Batch Normalization over an array."""
if x.ndim != 2:
batch_num = x.shape[0]
x = x.reshape(batch_num, -1)
@ -141,7 +141,7 @@ def _batch_norm_grad(dout, x, scale, save_mean, save_inv_variance, \
def batch_norm_grad(dy, x, scale, save_mean, save_inv_variance):
"""Batch normalization over an array."""
"""Batch Normalization over an array."""
if dy.ndim != 2:
batch_size = dy.shape[0]
dy = dy.reshape(batch_size, -1)
@ -275,7 +275,7 @@ def conv2d(x, weight, bias=None, stride=1, pad=0,
col = im2col(x, filter_h, filter_w, stride, pad, dilation)
col_w = np.reshape(weight, (filter_num, -1)).T
out = np.dot(col, col_w)
out = out.reshape(batch_num, out_h, out_w, -1).transpose(0, 3, 1, 2)
out = out.reshape((batch_num, out_h, out_w, -1)).transpose(0, 3, 1, 2)
if bias is not None:
out += bias
return out
@ -287,7 +287,7 @@ def conv2d_backprop_filter(dout, x, w_size, stride=1, pad=0):
dout = dout.transpose(0, 2, 3, 1).reshape(-1, filter_num)
col = im2col(x, filter_height, filter_width, stride, pad)
dw = np.dot(col.T, dout)
dw = dw.transpose(1, 0).reshape(filter_num, channel, filter_height, filter_width)
dw = dw.transpose(1, 0).reshape((filter_num, channel, filter_height, filter_width))
return dw