-
Notifications
You must be signed in to change notification settings - Fork 0
/
datasets.jl
59 lines (49 loc) · 1.49 KB
/
datasets.jl
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
module datasets
using Random
"""
Makes data with two classes.
Makes it as one big randomly shuffled minibatch.
Example shape (an unfilled square):
Square: Box:
0 0 0 0 0 0 0 0 0 0
0 1 1 1 0 0 1 1 1 0
0 1 0 1 0 0 1 1 1 0
0 1 1 1 0 0 1 1 1 0
0 0 0 0 0 0 0 0 0 0
For the labels, [1; 0] means its a square, and [0; 1] means its a box.
TODO: Make able to be inverted for the convolution
transpose case i.e. where the data is the target.
"""
function makeTwoClassShapes(boxDims::Tuple{Int64,Int64} = (5,5); isInvert::Bool = false)
any(x -> x < 3, boxDims) && throw(ArgumentError("each dimension of `boxDims` must be `>= 3`"))
r, c = boxDims
rSteps = r - 3 + 1
cSteps = c - 3 + 1
n = rSteps * cSteps * 2
data = zeros(Float32, r, c, 1, n)
labels = zeros(Float32, 2, n)
# Make the squares
i = 0
dataIndices = shuffle(1:n)
idx = 1
for sᵢ in 1:rSteps, sⱼ in 1:cSteps
i += 1
idx = dataIndices[i]
data[sᵢ, sⱼ:sⱼ+2, 1, idx] = [1. 1. 1.]
data[sᵢ+1, sⱼ:sⱼ+2, 1, idx] = [1. 0. 1.]
data[sᵢ+2, sⱼ:sⱼ+2, 1, idx] = [1. 1. 1.]
# Add the label
labels[1, idx] = 1
end
# make the boxes
for bᵢ in 1:rSteps, bⱼ in 1:cSteps
i += 1
idx = dataIndices[i]
data[bᵢ:bᵢ+2, bⱼ:bⱼ+2, 1, idx] = [1. 1. 1.; 1. 1. 1.; 1. 1. 1.;]
# Add the label
labels[2, idx] = 1
end
return (data, labels)
end
export makeTwoClassShapes
end # module datasets