a
h% @ s d dl mZ d dlmZmZmZ d dlZd dlmZmZ ddl m
Z
ddl m Z ddl
mZ d d
lmZmZmZ d d lmZ d d lmZmZmZ g d
ZG dd dejZG dd dejZ ddedZ G dd deZ e ede j fddddee e ee dd d Z!dS ) )partial)AnyCallableOptionalN)nnTensor )Conv2dNormActivation)ImageClassification)_log_api_usage_once )register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_make_divisible_ovewrite_named_paramhandle_legacy_interface) MobileNetV2MobileNet_V2_Weights mobilenet_v2c sL e Zd Zd eeeeeedejf dd fdd
Ze e dddZ
Z S )
InvertedResidualN.)inpoupstride expand_ratio
norm_layerreturnc s t || _|dvr&td| |d u r4tj}tt|| }| jdkoT||k| _g }|dkr~| t
||d|tj d | t
|||||tj dtj
||ddddd||g tj| | _|| _|dk| _d S ) N)r r z#stride should be 1 or 2 instead of r Z kernel_sizer activation_layer)r groupsr r r F)bias)super__init__r
ValueErrorr BatchNorm2dintrounduse_res_connectappendr ReLU6extendConv2d
SequentialconvZ out_channelsZ_is_cn)selfr r r r r Z
hidden_dimZlayers __class__ L/var/www/auris/lib/python3.9/site-packages/torchvision/models/mobilenetv2.pyr# s:
zInvertedResidual.__init__xr c C s" | j r|| | S | |S d S N)r( r. r/ r5 r2 r2 r3 forward< s zInvertedResidual.forward)N) __name__
__module__ __qualname__r& r r r Moduler# r r8
__classcell__r2 r2 r0 r3 r s (r c
sz e Zd Zdeeeeee eeedej f eedej f edd fdd
Z
e e d
d d Z e e d
d
dZ
ZS )r ?N 皙?.) num_classes
width_multinverted_residual_setting
round_nearestblockr dropoutr c s" t t| |du r t}|du r,tj}d}d} |du rjg dg dg dg dg dg d g d
g}t|d kst|d d krtd
| t|| |}t| t d| || _
t d|d|tj dg}
|D ]V\} } }
}t| | |}t
|
D ]2}|d kr|nd}|
||||| |d |}qq|
t || j
d|tj d tj|
| _ttj|dt| j
|| _| D ]}t|tjrtjj|jdd |jdurtj|j n`t|tjtj frtj |j tj|j n.t|tjrztj |jd d tj|j qzdS )aw
MobileNet V2 main class
Args:
num_classes (int): Number of classes
width_mult (float): Width multiplier - adjusts number of channels in each layer by this amount
inverted_residual_setting: Network structure
round_nearest (int): Round the number of channels in each layer to be a multiple of this number
Set to 1 to turn off rounding
block: Module specifying inverted residual building block for mobilenet
norm_layer: Module specifying the normalization layer to use
dropout (float): The droupout probability
N i )r r r ) r r )rJ rH r )rJ @ r )rJ ` rL r )rJ rL r )rJ i@ r r r rN zGinverted_residual_setting should be non-empty or a 4-element list, got r? rL r )r r r r )r r r )pZfan_out)modeg{Gz?) r" r# r r r r% lenr$ r max last_channelr r* ranger) r- featuresZDropoutZLinear
classifiermodules
isinstancer, initZkaiming_normal_Zweightr! Zzeros_Z GroupNormZones_Znormal_)r/ rB rC rD rE rF r rG Z
input_channelrU rW tcnsZoutput_channelir mr0 r2 r3 r# D sh
zMobileNetV2.__init__r4 c C s2 | |}tj|d}t|d}| |}|S )Nr r r )rW r Z
functionalZadaptive_avg_pool2dtorchflattenrX r7 r2 r2 r3
_forward_impl s
zMobileNetV2._forward_implc C s
| |S r6 )re r7 r2 r2 r3 r8 s zMobileNetV2.forward)r> r? Nr@ NNrA )r9 r: r; r& floatr listr r r< r# r re r8 r= r2 r2 r0 r3 r C s& _
r iz5 rb )Z
num_paramsZmin_size
categoriesc
@ sz e Zd Zedeeddi edddddid d
d d d
Zedeedddi edddddid ddd d
ZeZ dS )r z=https://download.pytorch.org/models/mobilenet_v2-b0353104.pth ) crop_sizezQhttps://github.com/pytorch/vision/tree/main/references/classification#mobilenetv2z ImageNet-1Kgx&1Q@gMV@)zacc@1zacc@5g$C?g\( +@zXThese weights reproduce closely the results of the paper using a simple training recipe.)ZrecipeZ_metricsZ_ops
_file_sizeZ_docs)urlZ
transformsmetaz=https://download.pytorch.org/models/mobilenet_v2-7ebf99e0.pth )rj Z resize_sizezHhttps://github.com/pytorch/vision/issues/3995#new-recipe-with-reg-tuningg`" R@gS㥛V@gV-2+@a$
These weights improve upon the results of the original paper by using a modified version of TorchVision's
`new training recipe
`_.
N)
r9 r: r; r r r
_COMMON_META
IMAGENET1K_V1Z
IMAGENET1K_V2DEFAULTr2 r2 r2 r3 r sF
r Z
pretrained)weightsT)rr progress)rr rs kwargsr c K sV t | } | dur(t|dt| jd tf i |}| durR|| j|dd |S )a MobileNetV2 architecture from the `MobileNetV2: Inverted Residuals and Linear
Bottlenecks `_ paper.
Args:
weights (:class:`~torchvision.models.MobileNet_V2_Weights`, optional): The
pretrained weights to use. See
:class:`~torchvision.models.MobileNet_V2_Weights` below for
more details, and possible values. By default, no pre-trained
weights are used.
progress (bool, optional): If True, displays a progress bar of the
download to stderr. Default is True.
**kwargs: parameters passed to the ``torchvision.models.mobilenetv2.MobileNetV2``
base class. Please refer to the `source code
`_
for more details about this class.
.. autoclass:: torchvision.models.MobileNet_V2_Weights
:members:
NrB rh T)rs Z
check_hash)r verifyr rS rm r Zload_state_dictZget_state_dict)rr rs rt modelr2 r2 r3 r s
r )" functoolsr typingr r r rc r r Zops.miscr Ztransforms._presetsr
utilsr Z_apir
r r Z_metar _utilsr r r __all__r< r r ro r rp boolr r2 r2 r2 r3 s0 0o*