mirror of
https://github.com/open-mmlab/mmclassification.git
synced 2025-06-03 21:53:55 +08:00
65410b05ad
* Refactor Mobilenetv3 structure and add ConvClsHead. * Change model's name from 'MobileNetv3' to 'MobileNetV3' * Modify configs for MobileNetV3 on CIFAR10. And add MobileNetV3 configs for imagenet * Fix activate setting bugs in MobileNetV3. And remove bias in SELayer. * Modify unittest * Remove useless config and file. * Fix mobilenetv3-large arch setting * Add dropout option in ConvClsHead * Fix MobilenetV3 structure according to torchvision version. 1. Remove with_expand_conv option in InvertedResidual, it should be decided by channels. 2. Revert activation function, should before SE layer. * Format code. * Rename MobilenetV3 arch "big" to "large". * Add mobilenetv3_small torchvision training recipe * Modify default `out_indices` of MobilenetV3, now it will change according to `arch` if not specified. * Add MobilenetV3 large config. * Add mobilenetv3 README * Modify InvertedResidual unit test. * Refactor ConvClsHead to StackedLinearClsHead, and add unit tests. * Add unit test for `simple_test` of `StackedLinearClsHead`. * Fix typo Co-authored-by: Yidi Shao <ydshao@smail.nju.edu.cn>
63 lines
2.2 KiB
Python
63 lines
2.2 KiB
Python
import mmcv
|
|
import torch.nn as nn
|
|
from mmcv.cnn import ConvModule
|
|
from mmcv.runner import BaseModule
|
|
|
|
from .make_divisible import make_divisible
|
|
|
|
|
|
# class SELayer(nn.Module):
|
|
class SELayer(BaseModule):
|
|
"""Squeeze-and-Excitation Module.
|
|
|
|
Args:
|
|
channels (int): The input (and output) channels of the SE layer.
|
|
ratio (int): Squeeze ratio in SELayer, the intermediate channel will be
|
|
``int(channels/ratio)``. Default: 16.
|
|
conv_cfg (None or dict): Config dict for convolution layer.
|
|
Default: None, which means using conv2d.
|
|
act_cfg (dict or Sequence[dict]): Config dict for activation layer.
|
|
If act_cfg is a dict, two activation layers will be configurated
|
|
by this dict. If act_cfg is a sequence of dicts, the first
|
|
activation layer will be configurated by the first dict and the
|
|
second activation layer will be configurated by the second dict.
|
|
Default: (dict(type='ReLU'), dict(type='Sigmoid'))
|
|
"""
|
|
|
|
def __init__(self,
|
|
channels,
|
|
ratio=16,
|
|
bias='auto',
|
|
conv_cfg=None,
|
|
act_cfg=(dict(type='ReLU'), dict(type='Sigmoid')),
|
|
init_cfg=None):
|
|
super(SELayer, self).__init__(init_cfg)
|
|
if isinstance(act_cfg, dict):
|
|
act_cfg = (act_cfg, act_cfg)
|
|
assert len(act_cfg) == 2
|
|
assert mmcv.is_tuple_of(act_cfg, dict)
|
|
self.global_avgpool = nn.AdaptiveAvgPool2d(1)
|
|
squeeze_channels = make_divisible(channels // ratio, 8)
|
|
self.conv1 = ConvModule(
|
|
in_channels=channels,
|
|
out_channels=squeeze_channels,
|
|
kernel_size=1,
|
|
stride=1,
|
|
bias=bias,
|
|
conv_cfg=conv_cfg,
|
|
act_cfg=act_cfg[0])
|
|
self.conv2 = ConvModule(
|
|
in_channels=squeeze_channels,
|
|
out_channels=channels,
|
|
kernel_size=1,
|
|
stride=1,
|
|
bias=bias,
|
|
conv_cfg=conv_cfg,
|
|
act_cfg=act_cfg[1])
|
|
|
|
def forward(self, x):
|
|
out = self.global_avgpool(x)
|
|
out = self.conv1(out)
|
|
out = self.conv2(out)
|
|
return x * out
|