U
|
©r_d ã @ s d Z ddddgZddlZddlZddlmZ ddlZddlm Z d d
|
l
|
m Z m Z G d d de ZG d d de ZG d d de ZG dd de ZdS )z9
|
@author: liaoxingyu
|
@contact: sherlockliao01@gmail.com
|
ÚToTensorÚ RandomErasingÚ RandomPatchÚAugMixé N)Údeque)ÚImageé )Ú to_tensorÚaugmentations_reidc @ s e Zd ZdZdd Zdd ZdS )r a Convert a ``PIL Image`` or ``numpy.ndarray`` to tensor.
|
|
Converts a PIL Image or numpy.ndarray (H x W x C) in the range
|
[0, 255] to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 255.0]
|
if the PIL Image belongs to one of the modes (L, LA, P, I, F, RGB, YCbCr, RGBA, CMYK, 1)
|
or if the numpy.ndarray has dtype = np.uint8
|
|
In the other cases, tensors are returned without scaling.
|
c C s t |S )z
|
Args:
|
pic (PIL Image or numpy.ndarray): Image to be converted to tensor.
|
|
Returns:
|
Tensor: Converted image.
|
)r )ÚselfÚpic© r ú ./data/transforms/transforms.pyÚ__call__ s zToTensor.__call__c C s | j jd S )Nz())Ú __class__Ú__name__)r r r r Ú__repr__( s zToTensor.__repr__N)r Ú
|
__module__Ú __qualname__Ú__doc__r r r r r r r s
|
|
c @ s0 e Zd ZdZdddddd fdd Zd
|
d Zd S ) r að Randomly selects a rectangle region in an image and erases its pixels.
|
'Random Erasing Data Augmentation' by Zhong et al.
|
See https://arxiv.org/pdf/1708.04896.pdf
|
Args:
|
probability: The probability that the Random Erasing operation will be performed.
|
sl: Minimum proportion of erased area against input image.
|
sh: Maximum proportion of erased area against input image.
|
r1: Minimum aspect ratio of erased area.
|
mean: Erasing value.
|
ç à?g{®Gáz?gÙ?g333333Ó?éÿ )g1*Ôß?gec]ÜÞ?gú~j¼tÜ?c C s" || _ || _|| _|| _|| _d S ©N)Ú probabilityÚmeanÚslÚshÚr1)r r r r r r r r r Ú__init__8 s
|
zRandomErasing.__init__c
|
C s t j|t jd ¡ }t dd¡| jkr*|S tdD ]^}|jd |jd }t | j | j
|
¡| }t | j d| j ¡}t t t || ¡}t t t || ¡}||jd k r2||jd k r2t d|jd | ¡}t d|jd | ¡} |jd dkrf| jd ||||
| | |
df< | jd ||||
| | |
df< | jd ||||
| | |
df< n$| jd ||||
| | |
df< | S q2|S )N©Údtyper r éd é é )ÚnpÚasarrayÚfloat32ÚcopyÚrandomÚuniformr ÚrangeÚshaper r r ÚintÚroundÚmathÚsqrtÚrandintr )
|
r ÚimgÚattemptÚareaÚ target_areaÚ aspect_ratioÚhÚwÚx1Úy1r r r r ? s& $$&$
|
zRandomErasing.__call__N©r r r r r r r r r r r , s c @ s2 e Zd ZdZdddZd d
|
Zd d Zd d ZdS )r ae Random patch data augmentation.
|
There is a patch pool that stores randomly extracted pathces from person images.
|
For each input image, RandomPatch
|
1) extracts a random patch and stores the patch in the patch pool;
|
2) randomly selects a patch from the patch pool and pastes it on the
|
input (at random position) to simulate occlusion.
|
Reference:
|
- Zhou et al. Omni-Scale Feature Learning for Person Re-Identification. ICCV, 2019.
|
- Zhou et al. Learning Generalisable Omni-Scale Representations
|
for Person Re-Identification. arXiv preprint, 2019.
|
r éPÃ r! ç{®Gáz?ç¹?c C s: || _ || _|| _|| _|| _|| _t|d| _|| _d S )N)Úmaxlen) Ú prob_happenÚpatch_min_areaÚpatch_max_areaÚpatch_min_ratioÚ prob_rotateÚprob_flip_leftrightr Ú patchpoolÚmin_sample_size) r r? Z pool_capacityrF r@ rA rB rC rD r r r r f s zRandomPatch.__init__c C s || }t dD ]t}t | j| j¡| }t | jd| j ¡}ttt || ¡}ttt || ¡}||k r||k r||f S qdS )Nr! g ð?)NN)
|
r* r( r) r@ rA rB r, r- r. r/ ) r ÚWÚHr3 r2 r4 r5 r6 r7 r r r Ú generate_whv s zRandomPatch.generate_whc C sF t dd¡| jkr | tj¡}t dd¡| jkrB| t dd¡¡}|S )Nr r iöÿÿÿé
|
) r( r) rD Ú transposer ÚFLIP_LEFT_RIGHTrC Úrotater0 )r Úpatchr r r Útransform_patch s
|
zRandomPatch.transform_patchc C s t |tjr t | tj¡¡}|j\}}| ||¡\}}|d k r|d k rt
|
d|| ¡}t
|
d|| ¡}| |||| || f¡}| j |¡ t| j | jk r¢|S t dd¡| jkr¸|S t | j d¡d } | j\}
|
} t
|
d||
|
¡}t
|
d|| ¡}| | ¡} | | ||f¡ |S )Nr r )Ú
|
isinstancer$ Úndarrayr Ú fromarrayÚastypeÚuint8ÚsizerI r( r0 ÚcroprE ÚappendÚlenrF r) r? ÚsamplerO Úpaste) r r1 rG rH r7 r6 r8 r9 Z new_patchrN ZpatchWZpatchHr r r r s(
|
|
|
zRandomPatch.__call__N)r r; r! r< r r= r r )r r r r r rI rO r r r r r r Y s þ
|
c @ s" e Zd ZdZd
|
ddZdd Zd S ) r a Perform AugMix augmentation and compute mixture.
|
Args:
|
aug_prob_coeff: Probability distribution coefficients.
|
mixture_width: Number of augmentation chains to mix per augmented example.
|
mixture_depth: Depth of augmentation chains. -1 denotes stochastic depth in [1, 3]'
|
severity: Severity of underlying augmentation operators (between 1 to 10).
|
r r# éÿÿÿÿc C s" || _ || _|| _|| _t| _d S r )Úaug_prob_coeffÚ mixture_widthÚ mixture_depthÚseverityr
|
Úaug_list)r r\ r] r^ r_ r r r r ° s
|
zAugMix.__init__c C s t t j | jg| j ¡¡}t t j | j| j¡¡}t j|t jd ¡ }t |¡}|j
|
d |j
|
d }}t | jD ]}t | ¡ t j¡¡}| jdkr| jn t j dd¡} t | D ]$}
|
t j | j¡} | || j||f}q²||| t j|t jd 7 }qrd| | || } | S )zwPerform AugMix augmentations and compute mixture.
|
Returns:
|
mixed: Augmented and mixed image.
|
r r r é )r$ r& r( Ú dirichletr\ r] Úbetar% r' Ú
|
zeros_liker+ r* r rR rS rT r^ r0 Úchoicer` r_ ) r ÚimageÚwsÚmZmixr6 r7 ÚiZ image_augÚdepthÚ_ÚopÚmixedr r r r · s ÿ
|
zAugMix.__call__N)r r# r[ r r: r r r r r § s
|
)r Ú__all__r. r( Ú collectionsr Únumpyr$ ÚPILr Ú
|
functionalr r
|
Úobjectr r r r r r r r Ú<module> s -N
|
