notebook.community

Edit and run



In [1]:

    
import os
import numpy as np
import sys
import cv2
import matplotlib.pyplot as plt
from sklearn.svm import NuSVC, SVC
import datetime



In [2]:

    
path = '/media/zhaoke/b0685ee4-63e3-4691-ae02-feceacff6996/data/'



In [3]:

    
paths = os.listdir(path)



In [4]:

    
len(paths)









    Out[4]:





160000



In [5]:

    
paths = [i for i in paths if '.txt' in i]



In [6]:

    
len(paths)









    Out[6]:





80000



In [7]:

    
t1 = paths[0]



In [8]:

    
f = open(path+t1, 'r')



In [9]:

    
lines = f.readlines()



In [10]:

    
lines









    Out[10]:





['536,292,609,301,606,323,533,313,1\n',
 '527,337,544,339,541,357,525,355,2\n',
 '638,351,658,353,655,370,636,367,3\n',
 '522,377,565,383,563,397,520,392,4\n',
 '605,388,622,390,620,405,603,402,5\n',
 '658,394,677,397,675,412,655,409,6\n',
 '514,417,746,446,739,492,507,463,7\n',
 '586,531,892,570,889,589,583,551,8\n']



In [11]:

    
lines1 = [i.replace('\n', '').split(',') for i in lines]



In [22]:

    
lines1 = np.array(lines1).astype(np.uint32)



In [23]:

    
lines1









    Out[23]:





array([[536, 292, 609, 301, 606, 323, 533, 313,   1],
       [527, 337, 544, 339, 541, 357, 525, 355,   2],
       [638, 351, 658, 353, 655, 370, 636, 367,   3],
       [522, 377, 565, 383, 563, 397, 520, 392,   4],
       [605, 388, 622, 390, 620, 405, 603, 402,   5],
       [658, 394, 677, 397, 675, 412, 655, 409,   6],
       [514, 417, 746, 446, 739, 492, 507, 463,   7],
       [586, 531, 892, 570, 889, 589, 583, 551,   8]], dtype=uint32)



In [24]:

    
boxes = np.empty((640000, 9))



In [10]:

    
cnt = 0
for txt in paths:
    f = open(path+txt, 'r')
    f.close()
    cnt += 1
    if cnt % 1000 == 0:
        print cnt



In [9]:

    
boxes = np.empty((640000, 9))
cnt = 0
for txt in paths:
    f = open(path+txt, 'r')
    lines = f.readlines()
    f.close()
    lines = [i.replace('\n', '').split(',') for i in lines]
    lines = np.array(lines).astype(np.uint32)
    boxes[cnt*8:cnt*8+8] = lines
    cnt += 1
    if cnt % 1000 == 0:
        print cnt



In [11]:

    
boxes.shape









    Out[11]:





(640000, 9)



In [12]:

    
sys.getsizeof(boxes)









    Out[12]:





46080112



In [13]:

    
clf = NuSVC()



In [14]:

    
start_time = datetime.datetime.now()
print start_time
clf.fit(boxes[:8000, :8], boxes[:8000, 8])
end_time = datetime.datetime.now()
print end_time - start_time









    



2017-10-23 19:50:54.741211
0:00:03.952691



In [25]:

    
start_time = datetime.datetime.now()
print start_time
clf.fit(boxes[:16000, :8], boxes[:16000, 8])
end_time = datetime.datetime.now()
print end_time - start_time









    



2017-10-23 14:01:18.177875
0:00:15.398890



In [26]:

    
start_time = datetime.datetime.now()
print start_time
clf.fit(boxes[:32000, :8], boxes[:32000, 8])
end_time = datetime.datetime.now()
print end_time - start_time









    



2017-10-23 14:01:38.790141
0:01:28.544892



In [31]:

    
clf.predict(boxes[10000:10008, :8])









    Out[31]:





array([ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.])



In [32]:

    
clf.predict(boxes[80000:80008, :8])









    Out[32]:





array([ 7.,  2.,  7.,  7.,  2.,  7.,  7.,  7.])



In [33]:

    
boxes









    Out[33]:





array([[ 536.,  292.,  609., ...,  533.,  313.,    1.],
       [ 527.,  337.,  544., ...,  525.,  355.,    2.],
       [ 638.,  351.,  658., ...,  636.,  367.,    3.],
       ..., 
       [ 570.,  303.,  592., ...,  569.,  327.,    6.],
       [ 416.,  360.,  669., ...,  414.,  433.,    7.],
       [ 507.,  529.,  837., ...,  506.,  558.,    8.]])



In [39]:

    
boxes_max = np.max(boxes[:, :8], axis=1)
boxes_min = np.min(boxes[:, :8], axis=1)



In [38]:

    
boxes[:8]









    Out[38]:





array([[ 536.,  292.,  609.,  301.,  606.,  323.,  533.,  313.,    1.],
       [ 527.,  337.,  544.,  339.,  541.,  357.,  525.,  355.,    2.],
       [ 638.,  351.,  658.,  353.,  655.,  370.,  636.,  367.,    3.],
       [ 522.,  377.,  565.,  383.,  563.,  397.,  520.,  392.,    4.],
       [ 605.,  388.,  622.,  390.,  620.,  405.,  603.,  402.,    5.],
       [ 658.,  394.,  677.,  397.,  675.,  412.,  655.,  409.,    6.],
       [ 514.,  417.,  746.,  446.,  739.,  492.,  507.,  463.,    7.],
       [ 586.,  531.,  892.,  570.,  889.,  589.,  583.,  551.,    8.]])



In [41]:

    
print boxes_max
print boxes_min









    



[ 609.  544.  658. ...,  592.  669.  837.]
[ 292.  337.  351. ...,  303.  360.  506.]



In [42]:

    
print boxes_max.shape
print boxes_min.shape









    



(640000,)
(640000,)



In [44]:

    
boxes_max = boxes_max.reshape((-1, 1))
boxes_min = boxes_min.reshape((-1, 1))



In [45]:

    
print boxes_max.shape
print boxes_min.shape









    



(640000, 1)
(640000, 1)



In [48]:

    
boxes_max - boxes_min









    Out[48]:





array([[ 317.],
       [ 207.],
       [ 307.],
       ..., 
       [ 289.],
       [ 309.],
       [ 331.]])



In [49]:

    
(boxes[:, :8] - boxes_min)/(boxes_max - boxes_min)









    Out[49]:





array([[ 0.76971609,  0.        ,  1.        , ...,  0.0977918 ,
         0.76025237,  0.06624606],
       [ 0.9178744 ,  0.        ,  1.        , ...,  0.09661836,
         0.90821256,  0.08695652],
       [ 0.93485342,  0.        ,  1.        , ...,  0.06188925,
         0.92833876,  0.05211726],
       ..., 
       [ 0.92387543,  0.        ,  1.        , ...,  0.08304498,
         0.92041522,  0.08304498],
       [ 0.18122977,  0.        ,  1.        , ...,  0.27508091,
         0.17475728,  0.23624595],
       [ 0.00302115,  0.0694864 ,  1.        , ...,  0.20241692,
         0.        ,  0.1570997 ]])



In [50]:

    
a = (boxes[:, :8] - boxes_min)/(boxes_max - boxes_min)



In [51]:

    
a.shape









    Out[51]:





(640000, 8)



In [66]:

    
clf = SVC()
start_time = datetime.datetime.now()
print start_time
clf.fit(a[:32000], boxes[:32000, 8])
end_time = datetime.datetime.now()
print end_time - start_time









    



2017-10-23 14:17:38.733661
0:00:33.870528



In [71]:

    
clf.predict(a[80016:80100])









    Out[71]:





array([ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.,  1.,  3.,  3.,  1.,  3.,
        3.,  1.,  8.,  1.,  3.,  3.,  3.,  3.,  3.,  1.,  8.,  1.,  2.,
        3.,  7.,  2.,  3.,  7.,  8.,  1.,  3.,  3.,  1.,  3.,  3.,  7.,
        8.,  1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.,  1.,  1.,  3.,  7.,
        6.,  3.,  7.,  8.,  1.,  3.,  3.,  3.,  3.,  3.,  1.,  8.,  1.,
        3.,  3.,  1.,  3.,  3.,  7.,  8.,  1.,  3.,  3.,  1.,  3.,  3.,
        7.,  8.,  1.,  2.,  3.,  1.])



In [68]:

    
clf.predict(a[10000:10008])









    Out[68]:





array([ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.])



In [72]:

    
clf = SVC()
start_time = datetime.datetime.now()
print start_time
clf.fit(a[:32000], boxes[:32000, 8])
end_time = datetime.datetime.now()
print end_time - start_time









    



2017-10-23 14:20:28.124333
0:00:34.372978



In [80]:

    
clf.predict(a[80016:80096]).reshape((-1, 8))









    Out[80]:





array([[ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.],
       [ 1.,  3.,  3.,  1.,  3.,  3.,  1.,  8.],
       [ 1.,  3.,  3.,  3.,  3.,  3.,  1.,  8.],
       [ 1.,  2.,  3.,  7.,  2.,  3.,  7.,  8.],
       [ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.],
       [ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.],
       [ 1.,  1.,  3.,  7.,  6.,  3.,  7.,  8.],
       [ 1.,  3.,  3.,  3.,  3.,  3.,  1.,  8.],
       [ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.],
       [ 1.,  3.,  3.,  1.,  3.,  3.,  7.,  8.]])



In [116]:

    
#calculate the area
def area(p):
    p = p.reshape((-1, 2))
    return 0.5 * abs(sum(x0*y1 - x1*y0 
                    for ((x0, y0), (x1, y1)) in segments(p)))
def segments(p):
    return zip(p, np.concatenate((p[1:], [p[0]])))



In [123]:

    
p = [[0, 4, 2, 2, 2, 0, 0, 0], [0, 2, 2, 2, 2, 0, 0, 0]]



In [124]:

    
area(np.array(p[0]))









    Out[124]:





6.0



In [125]:

    
p = np.array(p)



In [126]:

    
p









    Out[126]:





array([[0, 4, 2, 2, 2, 0, 0, 0],
       [0, 2, 2, 2, 2, 0, 0, 0]])



In [127]:

    
np.apply_along_axis(area, 1, p)









    Out[127]:





array([ 6.,  4.])



In [129]:

    
boxes[:, :8]









    Out[129]:





array([[ 536.,  292.,  609., ...,  323.,  533.,  313.],
       [ 527.,  337.,  544., ...,  357.,  525.,  355.],
       [ 638.,  351.,  658., ...,  370.,  636.,  367.],
       ..., 
       [ 570.,  303.,  592., ...,  327.,  569.,  327.],
       [ 416.,  360.,  669., ...,  445.,  414.,  433.],
       [ 507.,  529.,  837., ...,  573.,  506.,  558.]])



In [131]:

    
areas = np.apply_along_axis(area, 1, boxes[:, :8])



In [135]:

    
area0 = area(boxes[:, :8][0])



In [136]:

    
area0









    Out[136]:





1598.0



In [137]:

    
area1 = area(boxes[:, :8][1])



In [138]:

    
area1









    Out[138]:





302.0



In [134]:

    
areas









    Out[134]:





array([  1598. ,    302. ,    328. , ...,    517.5,  18493. ,   9585. ])



In [151]:

    
areas = areas.reshape((-1, 8))



In [152]:

    
areas









    Out[152]:





array([[  1598. ,    302. ,    328. , ...,    300. ,  10875. ,   6082.5],
       [  2194.5,    441. ,    430.5, ...,    414. ,  13914.5,   7056. ],
       [  1214. ,    241. ,    274. , ...,    125.5,   7626. ,   4645.5],
       ..., 
       [  1449. ,    409. ,    387.5, ...,    384. ,  15132. ,   5268.5],
       [  1865.5,    384. ,    364.5, ...,    344. ,  11440.5,   7328. ],
       [  2858.5,    542. ,    627.5, ...,    517.5,  18493. ,   9585. ]])



In [154]:

    
id_areas = areas[:, 7]



In [155]:

    
id_areas









    Out[155]:





array([ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ])



In [170]:

    
id_areas_tile = np.tile(id_areas, (8, 1))



In [171]:

    
id_areas_tile









    Out[171]:





array([[ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ],
       [ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ],
       [ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ],
       ..., 
       [ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ],
       [ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ],
       [ 6082.5,  7056. ,  4645.5, ...,  5268.5,  7328. ,  9585. ]])



In [172]:

    
id_areas_tile.shape









    Out[172]:





(8, 80000)



In [173]:

    
id_areas_tile = id_areas_tile.T



In [174]:

    
id_areas_tile









    Out[174]:





array([[ 6082.5,  6082.5,  6082.5, ...,  6082.5,  6082.5,  6082.5],
       [ 7056. ,  7056. ,  7056. , ...,  7056. ,  7056. ,  7056. ],
       [ 4645.5,  4645.5,  4645.5, ...,  4645.5,  4645.5,  4645.5],
       ..., 
       [ 5268.5,  5268.5,  5268.5, ...,  5268.5,  5268.5,  5268.5],
       [ 7328. ,  7328. ,  7328. , ...,  7328. ,  7328. ,  7328. ],
       [ 9585. ,  9585. ,  9585. , ...,  9585. ,  9585. ,  9585. ]])



In [175]:

    
id_areas_tile.shape









    Out[175]:





(80000, 8)



In [176]:

    
areas / id_areas_tile









    Out[176]:





array([[ 0.26272092,  0.04965064,  0.0539252 , ...,  0.04932182,
         1.78791615,  1.        ],
       [ 0.3110119 ,  0.0625    ,  0.0610119 , ...,  0.05867347,
         1.97200964,  1.        ],
       [ 0.26132817,  0.05187816,  0.05898181, ...,  0.02701539,
         1.64158863,  1.        ],
       ..., 
       [ 0.27503084,  0.0776312 ,  0.07355035, ...,  0.07288602,
         2.87216475,  1.        ],
       [ 0.25457151,  0.05240175,  0.04974072, ...,  0.04694323,
         1.5612036 ,  1.        ],
       [ 0.2982264 ,  0.05654669,  0.06546688, ...,  0.05399061,
         1.92936881,  1.        ]])



In [143]:

    
boxes[boxes[:, 8]==8]









    Out[143]:





array([[ 586.,  531.,  892., ...,  583.,  551.,    8.],
       [ 581.,  436.,  917., ...,  581.,  457.,    8.],
       [ 500.,  651.,  749., ...,  498.,  669.,    8.],
       ..., 
       [ 614.,  550.,  871., ...,  614.,  570.,    8.],
       [ 666.,  606.,  926., ...,  668.,  634.,    8.],
       [ 507.,  529.,  837., ...,  506.,  558.,    8.]])



In [145]:

    
print boxes[7]
print boxes[15]









    



[ 586.  531.  892.  570.  889.  589.  583.  551.    8.]
[ 581.  436.  917.  430.  917.  451.  581.  457.    8.]



In [146]:

    
a = np.array([[1], [2], [3]])



In [147]:

    
a









    Out[147]:





array([[1],
       [2],
       [3]])



In [148]:

    
a * 3









    Out[148]:





array([[3],
       [6],
       [9]])



In [150]:

    
np.tile(a, (1, 3))









    Out[150]:





array([[1, 1, 1],
       [2, 2, 2],
       [3, 3, 3]])



In [177]:

    
boxes.shape









    Out[177]:





(640000, 9)



In [2]:

    
def calc_xy(p0, p1, p2):
    cos = calc_cos(p0, p1, p2)
    dis = calc_dis(p0, p2)
    return dis * cos, dis * np.sqrt(1 - np.square(cos))

def calc_dis(p0, p1):
    return np.sqrt(np.sum(np.square(p0-p1)))

def calc_cos(p0, p1, p2):
    A = p1 - p0
    B = p2 - p0
    num = np.dot(A, B)
    demon = np.linalg.norm(A) * np.linalg.norm(B)
    return num / demon



In [181]:

    
test.shape









    Out[181]:





(640000, 18)



In [9]:

    
#calculate the area
def area(p):
    p = p.reshape((-1, 2))
    return 0.5 * abs(sum(x0*y1 - x1*y0 
                    for ((x0, y0), (x1, y1)) in segments(p)))
def segments(p):
    return zip(p, np.concatenate((p[1:], [p[0]])))



In [10]:

    
def calc_new_xy(boxes):
    box0 = boxes[0]
    box1 = boxes[1]
    x, y = calc_xy(box1[4:6], box1[6:], box0[:2])
    dis = calc_dis(box1[4:6], box1[6:])
    area0 = area(box0)
    area1 = area(box1)
    return x/dis, y/dis, area0/area1



In [ ]: