-
Notifications
You must be signed in to change notification settings - Fork 0
/
time_from_image.py
203 lines (164 loc) · 7.13 KB
/
time_from_image.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
from PIL import Image
import IPython.display
import os
import collections
import numpy as np
from sklearn import datasets, svm, metrics
import sklearn
class time_from_image:
def __init__(self):
self.classifier = None
self.total_samples = 0
self.all_images = None
self.all_digits = None
self.data_s = None
self.digits_s = None
self.digits = []
self.digit_hashes = []
self.random_state = 42
self.training_size = 0.8
self.directories = []
self.index_ranges = []
self.numbers = []
self.index_ranges.append((6740, 10**6))
self.directories.append('movies_images/Super Mario Kart (USA)-3.bk2/')
self.numbers.append([])
self.numbers[-1].append('01741')
self.numbers[-1].append('01534')
self.numbers[-1].append('01575')
self.numbers[-1].append('02093')
self.numbers[-1].append('01494')
self.numbers[-1].append('12437')
self.index_ranges.append((6559, 10**6))
self.directories.append('movies_images/Super Mario Kart (USA)-5.bk2/')
self.numbers.append([])
self.numbers[-1].append('01809')
self.numbers[-1].append('01660')
self.numbers[-1].append('01547')
self.numbers[-1].append('01464')
self.numbers[-1].append('01587')
self.numbers[-1].append('12067')
self.index_ranges.append((5601, 10**6))
self.directories.append('movies_images/Super Mario Kart (USA)-6.bk2/')
self.numbers.append([])
self.numbers[-1].append('01653')
self.numbers[-1].append('01477')
self.numbers[-1].append('01654')
self.numbers[-1].append('01522')
self.numbers[-1].append('01476')
self.numbers[-1].append('11782')
self.index_ranges.append((5788, 10**6))
self.directories.append('movies_images/Super Mario Kart (USA)-7.bk2/')
self.numbers.append([])
self.numbers[-1].append('01640')
self.numbers[-1].append('01480')
self.numbers[-1].append('02072')
self.numbers[-1].append('01467')
self.numbers[-1].append('01465')
self.numbers[-1].append('12124')
self.index_ranges.append((5746, 10**6))
self.directories.append('movies_images/Super Mario Kart (USA)-8.bk2/')
self.numbers.append([])
self.numbers[-1].append('01661')
self.numbers[-1].append('01634')
self.numbers[-1].append('01455')
self.numbers[-1].append('01758')
self.numbers[-1].append('01488')
self.numbers[-1].append('11996')
self.filenames = []
self.dig_pos = []
for i in range(5):
self.dig_pos.append((0, i * 9, 8, i * 9 + 8))
self.dig_pos.append((16, i * 9, 24, i * 9 + 8))
self.dig_pos.append((24, i * 9, 32, i * 9 + 8))
self.dig_pos.append((40, i * 9, 48, i * 9 + 8))
self.dig_pos.append((48, i * 9, 56, i * 9 + 8))
i = 52
self.dig_pos.append((0, i, 8, i + 8))
self.dig_pos.append((16, i, 24, i + 8))
self.dig_pos.append((24, i, 32, i + 8))
self.dig_pos.append((40, i, 48, i + 8))
self.dig_pos.append((48, i, 56, i + 8))
for movie, direc in enumerate(self.directories):
self.filenames.append([])
for a, b, c in os.walk(direc):
for filename in c:
if filename.endswith('.png'):
index = int(filename.split('_frame_')[1].split('.')[0])
if index >= (self.index_ranges[movie][0]) and index < self.index_ranges[movie][1]:
self.filenames[-1].append(os.path.join(direc, filename))
def image_arrays_from_file(self, filename, movie, numbers):
x = 112
y = 37
width = 60
height = 60
img = Image.open(filename).crop((x, y, x + width, y + height)).convert('LA')
digits = []
digit_hashes = []
sorted_images = []
for i in range(10):
digits.append([])
digit_hashes.append([])
for i, pos in enumerate(self.dig_pos):
digit_array = np.array(img.crop(pos))[:,:,0]
h = hash(digit_array.tostring())
n = int(numbers[movie][i // 5][i % 5])
if h not in self.digit_hashes[n]:
self.digits[n].append(digit_array)
self.digit_hashes[n].append(h)
digits[n].append(digit_array)
digit_hashes[n].append(h)
sorted_images.append(digit_array)
return digits, digit_hashes, sorted_images
def filenames_to_digits(self, filenames, numbers):
self.digits = []
self.digit_hashes = []
for i in range(10):
self.digits.append([])
self.digit_hashes.append([])
for movie, filename in enumerate(filenames):
for file in filename:
self.image_arrays_from_file(file, movie, numbers)
def predict_time_from_filenames(self, filenames):
if not self.classifier:
self.classify()
if isinstance(filenames, str):
filenames = [filenames]
solution = collections.defaultdict(int)
for filename in filenames:
x, xx, xxx = self.image_arrays_from_file(filename, i, self.numbers)
solution[''.join([str(s)[0] for s in list(self.classifier.predict(np.array(xxx[-5:]).reshape(5, -1)))])] += 1
return solution
def timestring_to_time(self, timestring):
t = 0
t += int(timestring[0]) * 60
t += int(timestring[1]) * 10
t += int(timestring[2])
t += int(timestring[3]) * 0.1
t += int(timestring[4]) * 0.01
return t
def test_classifier(self):
for i, filename in enumerate(self.filenames):
prediction = self.predict_time_from_filenames(filename)
best_score = max([int(i) for i in list(prediction)])
if best_score != int(self.numbers[i][5]):
print('failed')
print('Expected: {}\nGot:{}'.format(self.numbers[i][5], prediction))
print(i)
return False
return True
def classify(self):
self.classifier = sklearn.svm.SVC(gamma=0.001)
self.total_samples = 0
for i in range(10):
self.total_samples += len(self.digits[i])
self.all_images = np.zeros([self.total_samples, 8, 8])
self.all_digits = np.zeros([self.total_samples])
index = 0
for i in range(10):
for d in self.digits[i]:
self.all_images[index] = d
self.all_digits[index] = i
index += 1
self.data_s, self.digits_s = sklearn.utils.shuffle(self.all_images.reshape(self.total_samples, -1), self.all_digits, random_state=self.random_state)
self.classifier.fit(self.data_s[0:int(self.total_samples * self.training_size)], np.ravel(self.digits_s[0:int(self.total_samples * self.training_size)]))