leibniz 0.2.0

The package provides a differentiable vector graphics rasterization loss.
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

//! Linear winding contributions.

use ::burn::tensor::{Bool, Tensor, backend::Backend};

use super::super::roots;

/// Quadratic-free polynomial coefficients `(slope, intercept)` per segment.
pub type Coefficients<B, const D: usize> = (Tensor<B, D>, Tensor<B, D>);

pub fn evaluate<B: Backend, const D: usize>(
    x_coefficients: Coefficients<B, D>,
    y_coefficients: Coefficients<B, D>,
    x: Tensor<B, D>,
    y: Tensor<B, D>,
) -> Tensor<B, D> {
    let (x1, x0) = x_coefficients;
    let (y1, y0) = y_coefficients;
    let roots = roots::linear::solve(y1.clone(), y0 - y);
    let x_offset = x - x0;
    let sign = y1.greater_elem(0.0).float() * 2.0 - 1.0;

    root_contribution(roots.t, roots.valid, x_offset, x1, sign)
}

pub fn point_coefficients<B: Backend, const D: usize>(
    p0: Tensor<B, D>,
    p1: Tensor<B, D>,
) -> Coefficients<B, D> {
    (p1 - p0.clone(), p0)
}

fn root_contribution<B: Backend, const D: usize>(
    t: Tensor<B, D>,
    valid: Tensor<B, D, Bool>,
    x_offset: Tensor<B, D>,
    x_coefficient: Tensor<B, D>,
    sign: Tensor<B, D>,
) -> Tensor<B, D> {
    let curve_x = x_coefficient * t.clone();
    let valid = valid
        .bool_and(t.clone().greater_equal_elem(0.0))
        .bool_and(t.lower_equal_elem(1.0))
        .bool_and(curve_x.greater(x_offset));

    valid.float() * sign
}