-
Notifications
You must be signed in to change notification settings - Fork 2
/
jugglingdataloader.py
184 lines (166 loc) · 7.6 KB
/
jugglingdataloader.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
import csv
import cv2
import numpy as np
from random import shuffle
from keras.preprocessing.image import ImageDataGenerator
from keras.utils import Sequence
import random
class JugglingDataLoader(Sequence):
def __init__(self, shape=(64,64), timesteps=1, batch_size=8, gridShape=(15,15), expressFactor=1, imageGenerator=ImageDataGenerator(), dataType='BGR', nballs=[1,2,3]):
self.batch_size = batch_size
self.expressFactor = expressFactor
self.dataType = dataType
self.nballs = nballs
self.channels = 3
self.timesteps = timesteps
self.gridShape = gridShape
self.shape = shape
self.dataFolder = "../data/"
self.submovavg150 = "../submovavg150/"
self.annotationsFolder = self.dataFolder + "annotations/"
self.framesFolder = self.dataFolder + "frames/"
self.allRows = []
self.trainRows = self._loadSetRows("trainvideos", self.expressFactor)
self.validationRows = self._loadSetRows("validationvideos", self.expressFactor)
self.testRows = self._loadSetRows("testvideos")
self.timestepShuffle()
self.imageGenerator = imageGenerator
def __len__(self):
return len(self.trainRows) // self.batch_size
def __getitem__(self, idx):
images = np.zeros((self.batch_size,self.shape[0],self.shape[1],self.channels))
grids = np.zeros((self.batch_size,*self.gridShape,9))
for i in range(self.batch_size):
localDict = self.imageGenerator.get_random_transform((256,256))
row = self.trainRows[idx*self.batch_size+i]
images[i] = self.getImage(row[0], transDict=localDict)
grids[i] = self.getGrid(row, transDict=localDict)
return images, grids
def on_epoch_end(self):
self.timestepShuffle()
def _loadSetRows(self, filename, expressFactor=1):
setRows = []
with open(self.dataFolder + filename) as f:
for videoline in f:
videoline = videoline.rstrip('\n')
if int(videoline[0]) not in self.nballs:
continue
with open(self.annotationsFolder + videoline) as csvfile:
readCSV = csv.reader(csvfile, delimiter=',')
count = 0
for row in readCSV:
if count < 300 * expressFactor:
if not(self.dataType == "SUBMOVAVG" and count < 1):
setRows.append(row)
self.allRows.append(row)
count += 1
return setRows
def timestepShuffle(self, timesteps=False):
if timesteps == False:
timesteps = self.timesteps
tmpList = [self.trainRows[i:i+timesteps] for i in range(0, len(self.trainRows), timesteps)]
shuffle(tmpList)
self.trainRows = []
for lists in tmpList:
for row in lists:
self.trainRows.append(row)
def streamTrainSet(self):
for row in self.trainRows:
yield cv2.imread(self.framesFolder + row[0]), self.getCoordinates(row)
def streamValidationSet(self):
for row in self.validationRows:
yield cv2.imread(self.framesFolder + row[0]), self.getCoordinates(row)
def streamTestSet(self):
for row in self.testRows:
yield cv2.imread(self.framesFolder + row[0]), self.getCoordinates(row)
def streamAll(self):
for row in self.allRows:
yield cv2.imread(self.framesFolder + row[0]), self.getCoordinates(row)
def getImage(self, filename, transDict=None):
if self.dataType == 'BGR':
return self.getBGR(filename, transDict)
elif self.dataType == 'SUBMOVAVG':
return self.getSubMovAvg(filename, transDict)
assert False, 'Invalid choice of dataType.'
def getBGR(self, filename, transDict=None):
img = cv2.imread(self.framesFolder + filename)
if transDict != None:
img = self.transformImage(img, transDict)
img = cv2.resize(img, self.shape, cv2.INTER_CUBIC)
img = img - np.min(img)
img = img / np.max(img)
return img
def getSubMovAvg(self, filename, transDict=None):
img = cv2.imread(self.submovavg150 + filename)
if transDict != None:
img = self.transformImage(img, transDict)
img = img.astype(np.float32) / 256
img = cv2.resize(img, self.shape, cv2.INTER_CUBIC)
return img
def transformImage(self, img, transDict):
return self.imageGenerator.apply_transform(img, transform_parameters=transDict)
def getGrid(self, row, transDict=False):
coordinates = self.getCoordinates(row, transDict)
coordinates = coordinates / 256.
gridWidth = self.gridShape[0]
gridHeight = self.gridShape[1]
boxWidth = 1. / gridWidth
boxHeight = 1. / gridHeight
grid = np.zeros((gridWidth, gridHeight, 9))
for i in range(4, len(coordinates), 2):
xIndex = int(coordinates[i] // boxWidth)
yIndex = int(coordinates[i+1] // boxHeight)
if xIndex >= 0 and xIndex < gridWidth and yIndex >= 0 and yIndex < gridHeight:
grid[xIndex,yIndex,0] = 1
grid[xIndex,yIndex,1] = (coordinates[i] - xIndex*boxWidth) / boxWidth
grid[xIndex,yIndex,2] = (coordinates[i+1] - yIndex*boxHeight) / boxHeight
for i in range(0, 4, 2):
xIndex = int(coordinates[i] // boxWidth)
yIndex = int(coordinates[i+1] // boxHeight)
if xIndex >= 0 and xIndex < gridWidth and yIndex >= 0 and yIndex < gridHeight:
grid[xIndex,yIndex,3+3*i//2] = 1
grid[xIndex,yIndex,4+3*i//2] = (coordinates[i] - xIndex*boxWidth) / boxWidth
grid[xIndex,yIndex,5+3*i//2] = (coordinates[i+1] - yIndex*boxHeight) / boxHeight
return grid
def getCoordinates(self, row, transDict=False):
coordinates = []
nballs = int(row[0][0])
for b in range(nballs*2+4):
coordinates.append(int(row[1+b]))
coordinates = np.array(coordinates)
if transDict != False:
coordinates = self.transformCoordinates(coordinates, transDict)
return coordinates
def transformCoordinates(self, coordinates, transDict):
for c in range(len(coordinates) // 2):
coordinates[c*2+1] -= transDict["tx"]
coordinates[c*2] -= transDict["ty"]
coordinates[c*2+1] = (coordinates[c*2+1] -128 ) / transDict["zx"] + 128
coordinates[c*2] = (coordinates[c*2] -128) / transDict["zy"] + 128
if transDict["flip_horizontal"] == True:
coordinates[c*2] = 255 - coordinates[c*2]
if transDict["flip_horizontal"] == True:
tmpHand = coordinates[0:2]
coordinates[0:2] = coordinates[2:4]
coordinates[2:4] = tmpHand
return coordinates
def getValidationSet(self):
count = len(self.validationRows)
images = np.zeros((count*2,self.shape[0],self.shape[1],self.channels))
grids = np.zeros((count*2,*self.gridShape,9))
i = 0
for row in self.validationRows:
images[i] = self.getImage(row[0])
grids[i] = self.getGrid(row)
i = i + 1
transDict = {}
transDict["flip_horizontal"] = True
transDict["tx"] = 0
transDict["ty"] = 0
transDict["zx"] = 1
transDict["zy"] = 1
for row in self.validationRows:
images[i] = self.getImage(row[0], transDict)
grids[i] = self.getGrid(row, transDict)
i = i + 1
return images, grids