Function Differences with torch.nn.GroupNorm

torch.nn.GroupNorm

class torch.nn.GroupNorm(
    num_groups,
    num_channels,
    eps=1e-05,
    affine=True
)(input) -> Tensor

For more information, see torch.nn.GroupNorm.

mindspore.nn.GroupNorm

class mindspore.nn.GroupNorm(
    num_groups,
    num_channels,
    eps=1e-05,
    affine=True,
    gamma_init='ones',
    beta_init='zeros'
)(x) -> Tensor

For more information, see mindspore.nn.GroupNorm.

Differences

PyTorch: Group normalization is performed on the mini-batch input by dividing the channels into groups and then calculating the mean and variance within each group for normalization.

MindSpore: MindSpore API implements basically the same function as PyTorch. MindSpore can also perform additional initialization of the radiating parameters that need to be learned.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	num_groups	num_groups	-
	Parameter 2	num_channels	num_channels	-
	Parameter 3	eps	eps	-
	Parameter 4	affine	affine	-
	Parameter 5	-	gamma_init	Initialize the radial transform parameter gamma used for learning in the formula. The default is 'ones', while PyTorch cannot be set additionally, can only be 'ones'.
	Parameter 6	-	beta_init	Initialize the radial transform parameter beta used for learning in the formula. The default is 'ones', while PyTorch cannot be set additionally, can only be 'ones'.
Input	Single input	input	x	Interface input, same function, different parameter names

Code Example 1

MindSpore API basically implements the same function as TensorFlow, and MindSpore can also perform additional initialization of the two learning parameters.

# PyTorch
import torch
import numpy as np
from torch import tensor, nn

x = tensor(np.ones([1, 2, 4, 4], np.float32))
net = nn.GroupNorm(2, 2)
output = net(x).detach().numpy()
print(output)
# [[[[0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]]
#
#   [[0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]]]]

# MindSpore
import mindspore as ms
import numpy as np
from mindspore import Tensor, nn

x = Tensor(np.ones([1, 2, 4, 4], np.float32))
net = nn.GroupNorm(2, 2)
output = net(x)
print(output)
# [[[[0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]]
#
#   [[0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]
#    [0. 0. 0. 0.]]]]

MindSpore / docs

Function Differences with torch.nn.GroupNorm

torch.nn.GroupNorm

mindspore.nn.GroupNorm

Differences

Code Example 1

简介

发行版

贡献者

近期动态

MindSpore / docs .gitee-modal { width: 500px !important; }

Function Differences with torch.nn.GroupNorm

torch.nn.GroupNorm

mindspore.nn.GroupNorm

Differences

Code Example 1

简介

发行版

开源评估指数源自 OSS-Compass 评估体系，评估体系围绕以下三个维度对项目展开评估：

贡献者

近期动态

搜索帮助

MindSpore / docs