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

/*!
# 2D Operations

This module contains operations on tensor of shape [N, C, H, W] on each pixel.

 */

use tch::Tensor;

/**
Compute the vector dot product between two tensors for each pixel.

# Arguments
a: Tensor - The first tensor [N, C, H, W]
b: Tensor - The second tensor [N, C, H, W]

# Returns
Tensor - The dot product of the two tensors [N, 1 ,H, W]

# Example
```rust,no_run
# use tch::Tensor;
# use tch_utils::ops_2d::dot_product_2d;
let a = Tensor::of_slice(&[1.0, 2.0, 3.0, 4.0]).view([1, 2, 2, 1]);
let b = Tensor::of_slice(&[1.0, 2.0, 3.0, 4.0]).view([1, 2, 2, 1]);
let dot = dot_product_2d(&a, &b);
assert!(dot.equal(&Tensor::of_slice(&[10.0, 20.0]).view([1, 1, 2, 1])));
```
 */
pub fn dot_product_2d(a: &Tensor, b: &Tensor) -> Tensor {
  let typ = a.kind();
  (a * b).sum_dim_intlist(vec![1].as_slice(), true, typ)
}

/**
Normalize vectors in 2d pictures to unit length.

# Arguments
a: Tensor - The tensor to normalize [N, C, H, W]

# Returns
Tensor - The normalized tensor [N, C, H, W]
 */
pub fn normalize_2d(a: &Tensor) -> Tensor {
  a / (norm_2d(a) + 1e-8)
}

/**
Compute the norm of vectors in 2d pictures. Equivalent to `linalg_norm` but faster.

# Arguments
- t: Tensor - The tensor to compute the norm [N, C, H, W]

# Returns
Tensor - The norm of the vectors [N, 1, H, W]
 */
pub fn norm_2d(t: &Tensor) -> Tensor {
  let typ = t.kind();
  (t.pow_tensor_scalar(2.0))
    .sum_dim_intlist(vec![1].as_slice(), true, typ)
    .sqrt()
}

/**
Scale the of vectors in 2d pictures.

# Arguments
t: Tensor - The tensor to scale [N, C, H, W]
scale: &[f64] - The scale factors (lenght == C)

# Returns
Tensor - The scaled tensor [N, C, H, W]
 */
pub fn scale_2d(t: &Tensor, scale: &[f64]) -> Tensor {
  let typ = t.kind();
  let device = t.device();
  let scale = Tensor::of_slice(scale)
    .to_kind(typ)
    .to_device(device)
    .view([1, scale.len() as i64, 1, 1]);
  t * scale
}

/**
Rotate the of vectors in 2d pictures.

# Arguments
t: Tensor - The tensor to rotate [N, 2, H, W]
angle: f64 - The angle of rotation in radians

# Returns
Tensor - The rotated tensor [N, 2, H, W]
 */
pub fn rotate_2d(t: &Tensor, angle: f64) -> Tensor {
  let typ = t.kind();
  let device = t.device();
  let cos = angle.cos();
  let sin = angle.sin();
  let rotation = Tensor::of_slice(&[cos, -sin, sin, cos])
    .to_kind(typ)
    .to_device(device)
    .view([2, 2]);
  
  t.transpose(1,3).matmul(&rotation).transpose(1,3)
}

/**
Translate the of vectors in 2d pictures.

# Arguments
t: Tensor - The tensor to translate [N, C, H, W]
translation: &[f64] - The translation factors (lenght == C)

# Returns
Tensor - The translated tensor [N, C, H, W]
 */
pub fn translate_2d(t: &Tensor, translation: &[f64]) -> Tensor {
  let typ = t.kind();
  let device = t.device();
  let translation = Tensor::of_slice(translation)
    .to_kind(typ)
    .to_device(device)
    .view([1, translation.len() as i64, 1, 1]);
  t + translation
}