3
Á3W^ç  ã               @   sv   d dl mZ ddlmZ ddgZddiZddd	„ZG d
d„ dejƒZG dd„ dej	ƒZ
G dd„ dej	ƒZddd„ZdS )é    )Únné   )Úload_state_dict_from_urlÚMobileNetV2Úmobilenet_v2z=https://download.pytorch.org/models/mobilenet_v2-b0353104.pthNc             C   sB   |dkr|}t |t| |d  ƒ| | ƒ}|d|  k r>||7 }|S )aD  
    This function is taken from the original tf repo.
    It ensures that all layers have a channel number that is divisible by 8
    It can be seen here:
    https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
    :param v:
    :param divisor:
    :param min_value:
    :return:
    Né   gÍÌÌÌÌÌì?)ÚmaxÚint)ÚvZdivisorZ	min_valueZnew_v© r   úC/tmp/pip-build-7wfi2gtz/torchvision/torchvision/models/mobilenet.pyÚ_make_divisible   s    r   c                   s   e Zd Zd‡ fdd„	Z‡  ZS )Ú
ConvBNReLUé   r   c          
      sF   |d d }t t| ƒjtj||||||dd�tj|ƒtjdd�ƒ d S )Nr   r   F)ÚgroupsÚbiasT)Zinplace)Úsuperr   Ú__init__r   ÚConv2dÚBatchNorm2dZReLU6)ÚselfZ	in_planesZ
out_planesÚkernel_sizeÚstrider   Úpadding)Ú	__class__r   r   r   "   s
    
zConvBNReLU.__init__)r   r   r   )Ú__name__Ú
__module__Ú__qualname__r   Ú__classcell__r   r   )r   r   r   !   s   r   c                   s$   e Zd Z‡ fdd„Zdd„ Z‡  ZS )ÚInvertedResidualc          
      s¨   t t| ƒjƒ  || _|dks t‚tt|| ƒƒ}| jdko@||k| _g }|dkrd|jt	||dd�ƒ |j
t	||||d�tj||ddddd�tj|ƒgƒ tj|Ž | _d S )	Nr   r   )r   )r   r   r   F)r   )r   r   )r   r   r   r   ÚAssertionErrorr	   ÚroundÚuse_res_connectÚappendr   Úextendr   r   r   Ú
SequentialÚconv)r   ÚinpZoupr   Úexpand_ratioZ
hidden_dimZlayers)r   r   r   r   ,   s    zInvertedResidual.__init__c             C   s"   | j r|| j|ƒ S | j|ƒS d S )N)r"   r&   )r   Úxr   r   r   ÚforwardA   s    zInvertedResidual.forward)r   r   r   r   r*   r   r   r   )r   r   r   +   s   r   c                   s&   e Zd Zd	‡ fdd„	Zdd„ Z‡  ZS )
r   éè  ç      ð?Né   c          
      s  t t| ƒjƒ  t}d}d}|dkrlddddgddddgddd	dgdd
ddgddd	dgddd	dgddddgg}t|ƒdksˆt|d ƒdkr–tdj|ƒƒ‚t|| |ƒ}t|td|ƒ |ƒ| _	t
d	|dd�g}x`|D ]X\}	}
}}t|
| |ƒ}x<t|ƒD ]0}|dk�r|nd}|j|||||	d�ƒ |}qôW qÐW |jt
|| j	dd�ƒ tj|Ž | _tjtjdƒtj| j	|ƒƒ| _x¨| jƒ D ]œ}t|tjƒ�r¶tjj|jdd� |jdk	�rtjj|jƒ nZt|tjƒ�râtjj|jƒ tjj|jƒ n.t|tjƒ�rvtjj|jddƒ tjj|jƒ �qvW dS )a£  
        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
        é    i   Nr   é   é   é   r   r   é@   é   é`   é    i@  r   zIinverted_residual_setting should be non-empty or a 4-element list, got {}g      ð?)r   )r(   )r   gš™™™™™É?Zfan_out)Úmodeg{®Gáz„?)r   r   r   r   ÚlenÚ
ValueErrorÚformatr   r   Úlast_channelr   Úranger#   r   r%   ÚfeaturesZDropoutZLinearÚ
classifierÚmodulesÚ
isinstancer   ÚinitZkaiming_normal_Zweightr   Zzeros_r   Zones_Znormal_)r   Znum_classesZ
width_multZinverted_residual_settingZround_nearestÚblockZinput_channelr:   r<   ÚtÚcÚnÚsZoutput_channelÚir   Úm)r   r   r   r   I   sP    





zMobileNetV2.__init__c             C   s&   | j |ƒ}|jddgƒ}| j|ƒ}|S )Nr   r   )r<   Zmeanr=   )r   r)   r   r   r   r*   �   s    

zMobileNetV2.forward)r+   r,   Nr-   )r   r   r   r   r*   r   r   r   )r   r   r   H   s   DFTc             K   s,   t f |Ž}| r(ttd |d�}|j|ƒ |S )aC  
    Constructs a MobileNetV2 architecture from
    `"MobileNetV2: Inverted Residuals and Linear Bottlenecks" <https://arxiv.org/abs/1801.04381>`_.

    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        progress (bool): If True, displays a progress bar of the download to stderr
    r   )Úprogress)r   r   Ú
model_urlsZload_state_dict)Z
pretrainedrH   ÚkwargsÚmodelZ
state_dictr   r   r   r   ”   s    	

)N)FT)Ztorchr   Úutilsr   Ú__all__rI   r   r%   r   ÚModuler   r   r   r   r   r   r   Ú<module>   s   

L