Datasets
Models
Downloads: 1
159 lines (159 with data), 11.3 kB
B
Æeé8 ã @ sÚ d Z ddlZddlZddlmZ ddlmZ ddlZ ddl
Z
ddl m Z m
Z
dd Zd'dd Zd(ddZdd ZG dd dZd)ddZd*ddZd+dd Zd,d d Ze
ed d fd d!Ze
ed"d d#d$fd%d&ZdS )-z
Model validation metrics
é N)ÚPath)Ú TryExceptÚthreadedc C s* ddddg}| d d
d d
f | d¡S )Ng g¹?gÍÌÌÌÌÌì?é é )Úsum)ÚxÚw© r
ú8/home/iml/Desktop/Talha/yolov5_P3_ROI_P/utils/metrics.pyÚfitness s r ç©?c C sf t t| | d d d }t |d ¡}t || d | || d fd¡}tj|t |¡| ddS )Né r r éÿÿÿÿÚvalid)Úmode)ÚroundÚlenÚnpÚonesÚ concatenateÚconvolve)ÚyÚfÚnfÚpÚypr
r
r Úsmooth s "r FÚ.r
ç¼Ø²Ò<Ú c s t | ¡} | | || || } }}t j|dd\ }
jd } t ddd¡g } }
t | | jd f¡t | df¡t | df¡ }}}x:t D ],\}}||k} |
| }| ¡ }|dks|dkrÌqd| | d¡}| | d¡}||| }t j | || |dd
df dd||< ||| }t j | || |dd
df dd||< xlt
| jd D ]Z}t |dd
|f |dd
|f \|||f< }}|rh|dkrh|
t | ||¡¡ qhW qW d| | || | } fd d
|
¡ D }tt|}|rt| |
|t|| d | t| | t|| d |d
d t| |t|| d |dd t| |t|| d |dd t| d¡d ¡ } |dd
| f |dd
| f | dd
| f }}} ||
¡ } | || | ¡ } | | ||| | t¡fS )a Compute the average precision, given the recall and precision curves.
Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
# Arguments
tp: True positives (nparray, nx1 or nx10).
conf: Objectness value from 0-1 (nparray).
pred_cls: Predicted object classes (nparray).
target_cls: True object classes (nparray).
plot: Plot precision-recall curve at mAP@0.5
save_dir: Plot save directory
# Returns
The average precision as computed in py-faster-rcnn.
T)Ú
return_countsr r iè N)Úleftr c s g | ]\}}| kr|qS r
r
)Ú.0ÚkÚv)Úunique_classesr
r ú
<listcomp>S s z ap_per_class.<locals>.<listcomp>z PR_curve.pngz F1_curve.pngÚF1)Úylabelz P_curve.pngÚ Precisionz R_curve.pngÚRecallg¹?)r ÚargsortÚuniqueÚshapeÚlinspaceÚzerosÚ enumerater ÚcumsumÚinterpÚrangeÚ
compute_apÚappendÚitemsÚdictÚ
plot_pr_curver Ú
plot_mc_curver ÚmeanÚargmaxr ÚastypeÚint) ÚtpÚconfÚpred_clsÚ
target_clsÚplotÚsave_dirÚnamesÚepsÚprefixÚiÚntÚncÚpxÚpyÚapr ÚrÚciÚcÚn_lÚn_pÚfpcÚtpcÚrecallÚ precisionÚjÚmpreÚmrecÚf1Úfpr
)r& r Ú ap_per_class sF
4 * *0 4 r\ c C sÄ t dg| dgf¡}t dg|dgf¡}t t j t |¡¡¡}d}|dkrrt ddd¡}t t |||¡|¡}nHt |dd
|dd
k¡d }t ||d || ||d ¡}|||fS ) zú Compute the average precision, given the recall and precision curves
# Arguments
recall: The recall curve (list)
precision: The precision curve (list)
# Returns
Average precision, precision curve, recall curve
g g ð?r3 r r ée Nr )
r r ÚflipÚmaximumÚ
accumulater/ Útrapzr3 Úwherer )rU rV rY rX Úmethodr rM rH r
r
r r5 b s
"&r5 c @ s@ e Zd ZdddZdd Zdd Zed dd
dZdd ZdS )ÚConfusionMatrixç Ð?çÍÌÌÌÌÌÜ?c C s. t |d |d f¡| _|| _|| _|| _d S )Nr )r r0 ÚmatrixrJ r@ Ú iou_thres)ÚselfrJ r@ rh r
r
r Ú__init__ s zConfusionMatrix.__init__c C s |dkr:| ¡ }x$|D ] }| j| j|f d7 < qW dS ||dd
df | jk }|dd
df ¡ }|dd
df ¡ }t|dd
dd
f |dd
dd
f }t || jk¡}|d jd r t t
|d¡||d |d f dd
df fd¡ ¡ ¡ }|d jd dkrª||dd
df
¡ ddd
}|tj|dd
df dd d }||dd
df
¡ ddd
}|tj|dd
df dd d }n
t d
¡}|jd dk} | ¡ t ¡\}
} } xjt|D ]^\}
}|
|
k}| rt|dkr| j|| | |f d7 < n| j| j|f d7 < qÖW | r|x<t|D ]0\}
}t| |
ksH| j|| jf d7 < qHW dS ) ai
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
detections (Array[N, 6]), x1, y1, x2, y2, conf, class
labels (Array[M, 5]), class, x1, y1, x2, y2
Returns:
None, updates confusion matrix accordingly
Nr r r é r r T)Ú return_index)r é )r> rg rJ r@ Úbox_iouÚtorchrb rh r. ÚcatÚstackÚcpuÚnumpyr, r r- r0 Ú transposer= r1 r Úany)ri Ú
detectionsÚlabelsÚ
gt_classesÚgcÚdetection_classesÚiour ÚmatchesÚnÚm0Úm1Ú_rH rW Údcr
r
r Ú
process_batch s:
*<"""$
z ConfusionMatrix.process_batchc C s2 | j ¡ }| j d¡| }|d d
|d d
fS )Nr r )rg Údiagonalr )ri r? r[ r
r
r Útp_fp´ s
zConfusionMatrix.tp_fpu+ WARNING â ï¸ ConfusionMatrix plot failureTr r
c
C s8 dd l }| j|r(| j d¡ dd¡d nd }tj||dk < tjddddd\}}| jt | }} |j
|d k rrd
nd d d| k od
k n o| |k}
|
r¨|dg nd} t ¡ < t
d¡ |j|||dk ddidddd| | d
d¡ W d Q R X | d¡ | d¡ | d¡ |jt|d d d t |¡ d S ) Nr r r gÖ&è .>g{®Gázt?)é é T)ÚfigsizeÚ tight_layouté2 g ð?gé?)Ú
font_scaleéc Ú
backgroundÚautoÚignoreé Úsizeé ÚBluesz.2fg ) ÚaxÚannotÚ annot_kwsÚcmapÚfmtÚsquareÚvminÚ xticklabelsÚ yticklabels)r r r ÚTrueÚ PredictedzConfusion Matrixzconfusion_matrix.pngéú )Údpi)Úseabornrg r Úreshaper ÚnanÚpltÚsubplotsrJ r ÚsetÚwarningsÚcatch_warningsÚ simplefilterÚheatmapÚ
set_facecolorÚ
set_xlabelÚ
set_ylabelÚ set_titleÚsavefigr Úclose) ri Ú normalizerD rE ÚsnÚarrayÚfigr rJ Únnrw Ú
ticklabelsr
r
r rC º s2 &
zConfusionMatrix.plotc C s6 x0t | jd D ] }td tt| j| ¡ qW d S )Nr Ú )r4 rJ ÚprintÚjoinÚmapÚstrrg )ri rH r
r
r r· Ù s zConfusionMatrix.printN)re rf )Tr r
) Ú__name__Ú
__module__Ú __qualname__rj r r r rC r· r
r
r
r rd ~ s
. rd TçH¯¼ò×z>c % C sl |r| dd¡| dd¡ \}}} }
\} } }
}| d |
d |
d |d f\}}}}|| || || || f\}}}}| | | | | | | | f\}}}}nX| dd¡\}}}}| dd¡\}}}}|| || |¡ } }
|| || |¡ }
}| |¡| |¡ d¡| |¡| |¡ d¡ }| |
|
| | | } || } |sZ|sZ|rh| |¡| |¡ } | |¡| |¡ } |s|rL| d | d | } || | | d || | | d d }!|r@dtjd t |
| ¡t | |
¡ d¡ }"t ¡ |"|"| d| }#W d Q R X | |!| |"|# S | |!| S | | | }$| |$| |$ S | S )Nr r r r r )
ÚchunkÚclampÚminimumr_ ÚmathÚpiro ÚatanÚpowÚno_grad)%Úbox1Úbox2ÚxywhÚGIoUÚDIoUÚCIoUrF Úx1Úy1Úw1Úh1Úx2Úy2Úw2Úh2Úw1_Úh1_Úw2_Úh2_Úb1_x1Úb1_x2Úb1_y1Úb1_y2Úb2_x1Úb2_x2Úb2_y1Úb2_y2ÚinterÚunionr{ ÚcwÚchÚc2Úrho2r% ÚalphaÚc_arear
r
r Úbbox_iouÞ s8 *$$& ,0
ré c C sz | d¡ dd¡| d¡ dd¡ \}}\}}t ||¡t ||¡ d¡ d¡}||| d¡|| d¡ | | S )a]
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
box1 (Tensor[N, 4])
box2 (Tensor[M, 4])
Returns:
iou (Tensor[N, M]): the NxM matrix containing the pairwise
IoU values for every element in boxes1 and boxes2
r r r )Ú unsqueezer¿ ro ÚminÚmaxrÀ Úprod)rÇ rÈ rF Úa1Úa2Úb1Úb2rá r
r
r rn s .$rn c
C sr | \}}}}|j \}}} }
t || ¡t ||¡ d¡t ||
¡t ||¡ d¡ } | | |
| | } | | S )zà Returns the intersection over box2 area given box1, box2. Boxes are x1y1x2y2
box1: np.array of shape(4)
box2: np.array of shape(nx4)
returns: np.array of shape(n)
r )ÚTr rÁ r_ Úclip)
rÇ rÈ rF rÙ rÛ rÚ rÜ rÝ rß rÞ rà Ú
inter_areaÚ box2_arear
r
r Úbbox_ioa s rö c C sJ | d d
d f } |d }t | |¡ d¡}|| d¡| d¡ | | S )Nr )ro rë rí )Zwh1Zwh2rF rá r
r
r Úwh_iou2 s r÷ z pr_curve.pngc
C s$ t jddddd\}}tj|dd}dt| k r<dk rn nDxTt|jD ]2\}}|j| |d|| d||df d d
qLW n|j| |dd d |j| | d¡d
dd|d d
df ¡ d | d¡ |
d¡ | dd¡ | dd¡ |j
ddd | d¡ |j|dd t |¡ d S )Nr )r é T)r r )Úaxisr é r¶ z.3f)Ú linewidthÚlabelÚgrey)rû Úcolorrm Úbluezall classes %.3f mAP@0.5)rû rþ rü r+ r* )g¤p=
×£ð?r z
upper left)Úbbox_to_anchorÚloczPrecision-Recall Curver )r )r£ r¤ r rq r r1 rò rC r; r« r¬ Úset_xlimÚset_ylimÚlegendr r® r¯ ) rK rL rM rD rE r³ r rH r r
r
r r9 = s 0.
r9 z mc_curve.pngÚ
ConfidenceÚMetricc
C s t jddddd\}}dt| k r.dk rbn n0xBt|D ] \}} |j| | d|| d q<W n|j| |jddd t| d¡d
} |j| | d d d
| ¡ dd| | ¡ dd |
|¡ | |¡ | dd¡ |
dd¡ |jddd | | d¡ |j|dd t |¡ d S )Nr )r rø T)r r r rú )rû rü rý )rû rþ g©?rm rÿ z all classes z.2fz at z.3f)rû rþ rü )g¤p=
×£ð?r z
upper left)r r z-Confidence Curver )r )r£ r¤ r r1 rC rò r r; rì r< r« r¬ r r r r r® r¯ )
rK rL rD rE Úxlabelr) r³ r rH r r
r
r r: T s 0
r: )r
)Fr r
r r )TFFFr¾ )r¾ )r¾ )r¾ )Ú__doc__r r¦ Úpathlibr Úmatplotlib.pyplotÚpyplotr£ rs r ro Úutilsr r r r r\ r5 rd ré rn rö r÷ r9 r: r
r
r
r Ú<module> s(
C `
)