U
    (d?.                     @   s   d dl Z d dlmZmZmZmZmZ d dlZd dlZd dlm	Z	m
Z
 ddlmZ ddlmZ ejje
e
ddd	Zejje
ed
ddZde
eeeeee
f  eeeef  ee
eeee
f  f dddZG dd de	jZe
ee ee e
dddZe
ee ee e
dddZdS )    N)ListTupleDictOptionalAny)nnTensor   )	ImageList)paste_masks_in_imageimagereturnc                 C   s   ddl m} || dd  S )Nr   )	operators)Z
torch.onnxr   Zshape_as_tensor)r   r    r   J/tmp/pip-unpacked-wheel-vx7f76es/torchvision/models/detection/transform.py_get_shape_onnx   s    r   )vr   c                 C   s   | S )Nr   )r   r   r   r   _fake_cast_onnx   s    r   )r   self_min_sizeself_max_sizetarget
fixed_sizer   c                 C   s0  t  rt| }nt| jdd  }d }d }d }|d k	rL|d |d g}nZt|jtjd}	t	|jtjd}
t||	 ||
 }t  rt
|}n| }d}tjjj| d  ||d|ddd } |d kr| |fS d	|kr(|d	 }tjjj|d d d f  |||d
d d df  }||d	< | |fS )Nr   r	   r   )dtypeTZbilinearF)sizescale_factormoderecompute_scale_factorZalign_cornersmasks)r   r   r   )torchvision_is_tracingr   torchtensorshapemintofloat32maxr   itemr   
functionalZinterpolatefloatbyte)r   r   r   r   r   Zim_shaper   r   r   min_sizemax_sizeZscalemaskr   r   r   _resize_image_and_masks   sN    

	
   

r0   c                
       s  e Zd ZdZd!eeee ee eeeeef  e	d fddZ
d"ee eeeeef   eeeeeeef   f ddd	Zeed
ddZee edddZd#eeeeef  eeeeeef  f dddZejjd$ee eedddZeee  ee dddZd%ee eedddZeeeef  eeeef  eeeef  eeeef  dddZeddd Z  ZS )&GeneralizedRCNNTransformag  
    Performs input / target transformation before feeding the data to a GeneralizedRCNN
    model.

    The transformations it perform are:
        - input normalization (mean subtraction and std division)
        - input / target resizing to match min_size / max_size

    It returns a ImageList for the inputs, and a List[Dict[Tensor]] for the targets
        N)r-   r.   
image_mean	image_stdsize_divisibler   kwargsc                    sT   t    t|ttfs|f}|| _|| _|| _|| _|| _	|| _
|dd| _d S )N_skip_resizeF)super__init__
isinstancelisttupler-   r.   r3   r4   r5   r   popr7   )selfr-   r.   r3   r4   r5   r   r6   	__class__r   r   r9   V   s    

z!GeneralizedRCNNTransform.__init__)imagestargetsr   c                 C   sB  dd |D }|d k	rPg }|D ],}i }|  D ]\}}|||< q.|| q|}tt|D ]v}|| }	|d k	rx|| nd }
|	 dkrtd|	j | |	}	| |	|
\}	}
|	||< |d k	r\|
d k	r\|
||< q\dd |D }| j	|| j
d}g }|D ]4}tt|dkd|  ||d	 |d
 f qt||}||fS )Nc                 S   s   g | ]}|qS r   r   .0imgr   r   r   
<listcomp>n   s     z4GeneralizedRCNNTransform.forward.<locals>.<listcomp>   zFimages is expected to be a list of 3d tensors of shape [C, H, W], got c                 S   s   g | ]}|j d d qS )r   Nr$   rC   r   r   r   rF      s     )r5      zMInput tensors expected to have in the last two elements H and W, instead got r   r	   )itemsappendrangelendim
ValueErrorr$   	normalizeresizebatch_imagesr5   r"   Z_assertr
   )r>   rA   rB   Ztargets_copytdatakr   ir   Ztarget_indexZimage_sizesZimage_sizes_listZ
image_size
image_listr   r   r   forwardk   s<    




z GeneralizedRCNNTransform.forwardr   c                 C   st   |  std|j d|j|j }}tj| j||d}tj| j||d}||d d d d f  |d d d d f  S )NzOExpected input images to be of floating type (in range [0, 1]), but found type z insteadr   device)Zis_floating_point	TypeErrorr   rZ   r"   Z	as_tensorr3   r4   )r>   r   r   rZ   ZmeanZstdr   r   r   rP      s    z"GeneralizedRCNNTransform.normalize)rU   r   c                 C   s*   t tddtt| }|| S )z
        Implements `random.choice` via torch ops so it can be compiled with
        TorchScript. Remove if https://github.com/pytorch/pytorch/issues/25803
        is fixed.
        r	   g        )intr"   emptyZuniform_r+   rM   r)   )r>   rU   indexr   r   r   torch_choice   s    "z%GeneralizedRCNNTransform.torch_choice)r   r   r   c                 C   s   |j dd  \}}| jr8| jr&||fS t| | j}nt| jd }t||t| j|| j\}}|d krr||fS |d }t	|||f|j dd  }||d< d|kr|d }t
|||f|j dd  }||d< ||fS )Nr   boxes	keypoints)r$   trainingr7   r+   r_   r-   r0   r.   r   resize_boxesresize_keypoints)r>   r   r   hwr   Zbboxrb   r   r   r   rQ      s"    zGeneralizedRCNNTransform.resize)rA   r5   r   c           
         s  g }t |d  D ]< tt fdd|D tjtj}|| q|}t	|d tj| | tj|d< t	|d tj| | tj|d< t
|}g }|D ]P}dd t|t
|jD }tjj|d|d d|d d|d f}	||	 qt|S )Nr   c                    s   g | ]}|j   qS r   rH   rC   rV   r   r   rF      s     z?GeneralizedRCNNTransform._onnx_batch_images.<locals>.<listcomp>r	   rI   c                 S   s   g | ]\}}|| qS r   r   )rD   s1s2r   r   r   rF      s     )rL   rN   r"   r(   stackr&   r'   Zint64rK   ceilr<   zipr$   r   r*   pad)
r>   rA   r5   r.   Z
max_size_istrideZpadded_imgsrE   paddingZ
padded_imgr   rh   r   _onnx_batch_images   s    .**(z+GeneralizedRCNNTransform._onnx_batch_images)the_listr   c                 C   sB   |d }|dd  D ](}t |D ]\}}t|| |||< q q|S )Nr   r	   )	enumerater(   )r>   rr   ZmaxesZsublistr^   r)   r   r   r   max_by_axis   s
    z$GeneralizedRCNNTransform.max_by_axisc           	      C   s   t  r| ||S | dd |D }t|}t|}ttt|d | | |d< ttt|d | | |d< t	|g| }|d 
|d}t|jd D ]@}|| }||d |jd d |jd d |jd f | q|S )Nc                 S   s   g | ]}t |jqS r   )r;   r$   rC   r   r   r   rF      s     z9GeneralizedRCNNTransform.batch_images.<locals>.<listcomp>r	   rI   r   )r    r!   rq   rt   r+   r;   r\   mathrl   rM   Znew_fullrL   r$   Zcopy_)	r>   rA   r5   r.   ro   Zbatch_shapeZbatched_imgsrV   rE   r   r   r   rR      s    ""6z%GeneralizedRCNNTransform.batch_images)resultimage_shapesoriginal_image_sizesr   c                 C   s   | j r
|S tt|||D ]~\}\}}}|d }t|||}||| d< d|krp|d }	t|	||}	|	|| d< d|kr|d }
t|
||}
|
|| d< q|S )Nra   r   rb   )rc   rs   rm   rd   r   re   )r>   rv   rw   rx   rV   predZim_sZo_im_sra   r   rb   r   r   r   postprocess   s    z$GeneralizedRCNNTransform.postprocess)r   c                 C   sZ   | j j d}d}|| d| j d| j d7 }|| d| j d| j d7 }|d	7 }|S )
N(z
    zNormalize(mean=z, std=)zResize(min_size=z, max_size=z, mode='bilinear')z
))r@   __name__r3   r4   r-   r.   )r>   format_string_indentr   r   r   __repr__  s    z!GeneralizedRCNNTransform.__repr__)r2   N)N)N)r2   )r2   )r}   
__module____qualname____doc__r\   r   r+   r   r   r   r9   r   r   strr
   rX   rP   r_   rQ   r"   jitunusedrq   rt   rR   rz   r   __classcell__r   r   r?   r   r1   J   sH       ) r1   )rb   original_sizenew_sizer   c           	         s    fddt ||D }|\}}  }tj r|d d d d df | }|d d d d df | }tj|||d d d d df fdd}n |d  |9  < |d  |9  < |S )	Nc                    s8   g | ]0\}}t j|t j jd t j|t j jd  qS rY   r"   r#   r'   rZ   rD   sZs_origrb   r   r   rF     s   z$resize_keypoints.<locals>.<listcomp>r   r	   rI   rN   ).r   ).r	   )rm   cloner"   Z_CZ_get_tracing_staterk   )	rb   r   r   ratiosZratio_hZratio_wZresized_dataZresized_data_0Zresized_data_1r   r   r   re     s    

(re   )ra   r   r   r   c           
         sh    fddt ||D }|\}} d\}}}}	|| }|| }|| }|	| }	tj||||	fddS )Nc                    s8   g | ]0\}}t j|t j jd t j|t j jd  qS r   r   r   ra   r   r   rF   )  s   z resize_boxes.<locals>.<listcomp>r	   r   )rm   Zunbindr"   rk   )
ra   r   r   r   Zratio_heightZratio_widthZxminZyminZxmaxZymaxr   r   r   rd   (  s    
rd   )NN)ru   typingr   r   r   r   r   r"   r    r   r   rW   r
   Z	roi_headsr   r   r   r   r+   r   r   r\   r0   Moduler1   re   rd   r   r   r   r   <module>   s0   	  1 M