embedded-nn 0.2.1

A pure Rust, #![no_std] neural network inference library inspired by CMSIS-NN for microcontrollers and embedded targets.
Documentation 
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

use embedded_nn::{
    pooling::{avg_pool_s16, avg_pool_s8, max_pool_s16, max_pool_s8},
    Activation, Dims, PoolParams, Tile,
};

#[test]
fn test_max_pool_s8_variations() {
    let pool_params = PoolParams {
        stride: Tile::new(2, 2),
        padding: Tile::new(0, 0),
        activation: Activation::new(-10, 10),
    };
    let filter_dims = Tile::new(2, 2);

    let input_dims = Dims::new(1, 4, 4, 1);
    let input = [
        -20i8, -15i8, 5i8, 6i8, -10i8, -5i8, 7i8, 8i8, 9i8, 10i8, 13i8, 14i8, 11i8, 12i8, 15i8,
        16i8,
    ];
    let output_dims = Dims::new(1, 2, 2, 1);
    let mut output = [0i8; 4];

    max_pool_s8(
        &pool_params,
        &filter_dims,
        &input_dims,
        &input,
        &output_dims,
        &mut output,
    )
    .unwrap();

    // Top-left block max = -5
    // Top-right block max = 8
    // Bottom-left block max = 12 -> clamped to 10
    // Bottom-right block max = 16 -> clamped to 10
    assert_eq!(output, [-5, 8, 10, 10]);
}

#[test]
fn test_avg_pool_s8_negative_values_and_padding() {
    let pool_params = PoolParams {
        stride: Tile::new(1, 1),
        padding: Tile::new(1, 1),
        activation: Activation::int8_unconstrained(),
    };
    let filter_dims = Tile::new(3, 3);

    let input_dims = Dims::new(1, 2, 2, 1);
    let input = [-10i8, -20i8, -30i8, -40i8];
    let output_dims = Dims::new(1, 2, 2, 1);
    let mut output = [0i8; 4];

    avg_pool_s8(
        &pool_params,
        &filter_dims,
        &input_dims,
        &input,
        &output_dims,
        &mut output,
    )
    .unwrap();

    // Top-left (0,0): covers 2x2 elements (-10, -20, -30, -40) -> sum = -100, count = 4
    // avg = -100 / 4 = -25
    assert_eq!(output[0], -25);
}

#[test]
fn test_max_pool_s16_multi_channel() {
    let pool_params = PoolParams {
        stride: Tile::new(1, 1),
        padding: Tile::new(0, 0),
        activation: Activation::int16_unconstrained(),
    };
    let filter_dims = Tile::new(2, 2);

    let input_dims = Dims::new(1, 2, 2, 2); // N=1, H=2, W=2, C=2
    let input = [
        100i16, -200i16, // (0,0)
        300i16, -400i16, // (0,1)
        500i16, -600i16, // (1,0)
        700i16, -100i16, // (1,1)
    ];
    let output_dims = Dims::new(1, 1, 1, 2);
    let mut output = [0i16; 2];

    max_pool_s16(
        &pool_params,
        &filter_dims,
        &input_dims,
        &input,
        &output_dims,
        &mut output,
    )
    .unwrap();

    // Ch 0 max: max(100, 300, 500, 700) = 700
    // Ch 1 max: max(-200, -400, -600, -100) = -100
    assert_eq!(output, [700, -100]);
}

#[test]
fn test_avg_pool_s16_large_sums() {
    let pool_params = PoolParams {
        stride: Tile::new(2, 2),
        padding: Tile::new(0, 0),
        activation: Activation::int16_unconstrained(),
    };
    let filter_dims = Tile::new(2, 2);

    let input_dims = Dims::new(1, 2, 2, 1);
    let input = [30000i16, 30000i16, 30000i16, 30000i16];
    let output_dims = Dims::new(1, 1, 1, 1);
    let mut output = [0i16; 1];

    avg_pool_s16(
        &pool_params,
        &filter_dims,
        &input_dims,
        &input,
        &output_dims,
        &mut output,
    )
    .unwrap();

    // sum = 120000 -> count = 4 -> avg = 30000
    assert_eq!(output[0], 30000);
}