loftr 0.1.1

Native Rust/tch implementation of LoFTR feature matching
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
130
131
132
133
134
135
136
137

use super::*;
use tch::{Device, Kind, nn};

fn cpu_module(config: &LoftrConfig) -> Result<FinePreprocess, LoftrError> {
    let vs = nn::VarStore::new(Device::Cpu);
    FinePreprocess::new(&vs.root(), config)
}

#[test]
fn empty_matches_return_empty_windows() -> Result<(), LoftrError> {
    let config = LoftrConfig::outdoor();
    let module = cpu_module(&config)?;
    let data = FinePreprocessData {
        hw0_f: (4, 4),
        hw0_c: (2, 2),
        b_ids: Tensor::zeros([0], (Kind::Int64, Device::Cpu)),
        i_ids: Tensor::zeros([0], (Kind::Int64, Device::Cpu)),
        j_ids: Tensor::zeros([0], (Kind::Int64, Device::Cpu)),
    };
    let fine_map0 = Tensor::zeros([1, 128, 4, 4], (Kind::Float, Device::Cpu));
    let fine_map1 = Tensor::zeros([1, 128, 4, 4], (Kind::Float, Device::Cpu));
    let coarse_tokens0 = Tensor::zeros([1, 4, 256], (Kind::Float, Device::Cpu));
    let coarse_tokens1 = Tensor::zeros([1, 4, 256], (Kind::Float, Device::Cpu));

    let (out0, out1) = module.forward(
        &fine_map0,
        &fine_map1,
        &coarse_tokens0,
        &coarse_tokens1,
        &data,
    )?;
    assert_eq!(out0.size(), vec![0, 25, 128]);
    assert_eq!(out1.size(), vec![0, 25, 128]);
    Ok(())
}

#[test]
fn unfold_local_windows_matches_expected_patch_layout() -> Result<(), LoftrError> {
    let unfolded = unfold_local_windows(
        &Tensor::arange_start(1, 17, (Kind::Float, Device::Cpu)).view([1, 1, 4, 4]),
        3,
        2,
    )?;
    assert_eq!(unfolded.size(), vec![1, 4, 9, 1]);

    let top_left = unfolded.get(0).get(0).squeeze_dim(-1);
    let values: Vec<f32> = Vec::<f32>::try_from(top_left)?;
    for (actual, expected) in values
        .into_iter()
        .zip([0.0_f32, 0.0, 0.0, 0.0, 1.0, 2.0, 0.0, 5.0, 6.0])
    {
        assert!((actual - expected).abs() < f32::EPSILON);
    }
    Ok(())
}

#[test]
fn forward_selects_only_requested_windows_without_concat() -> Result<(), LoftrError> {
    let mut config = LoftrConfig::outdoor();
    config.fine_window_size = 3;
    config.fine_concat_coarse_feat = false;
    let module = cpu_module(&config)?;
    let data = FinePreprocessData {
        hw0_f: (4, 4),
        hw0_c: (2, 2),
        b_ids: Tensor::from_slice(&[0_i64, 0]),
        i_ids: Tensor::from_slice(&[0_i64, 3]),
        j_ids: Tensor::from_slice(&[1_i64, 2]),
    };
    let fine_map0 =
        Tensor::arange_start(1, 1 + 128 * 4 * 4, (Kind::Float, Device::Cpu)).view([1, 128, 4, 4]);
    let fine_map1 = (&fine_map0 + 10_000.0).shallow_clone();
    let coarse_tokens0 = Tensor::zeros([1, 4, 256], (Kind::Float, Device::Cpu));
    let coarse_tokens1 = Tensor::zeros([1, 4, 256], (Kind::Float, Device::Cpu));

    let (out0, out1) = module.forward(
        &fine_map0,
        &fine_map1,
        &coarse_tokens0,
        &coarse_tokens1,
        &data,
    )?;
    assert_eq!(out0.size(), vec![2, 9, 128]);
    assert_eq!(out1.size(), vec![2, 9, 128]);

    let first_center = out0.get(0).get(4).double_value(&[0]);
    let second_center = out0.get(1).get(4).double_value(&[0]);
    let first_right_center = out1.get(0).get(4).double_value(&[0]);
    let second_right_center = out1.get(1).get(4).double_value(&[0]);

    assert!((first_center - 1.0).abs() < f64::EPSILON);
    assert!((second_center - 11.0).abs() < f64::EPSILON);
    assert!((first_right_center - 10_003.0).abs() < f64::EPSILON);
    assert!((second_right_center - 10_009.0).abs() < f64::EPSILON);
    Ok(())
}

#[test]
fn forward_with_concat_keeps_output_shape() -> Result<(), LoftrError> {
    let config = LoftrConfig::outdoor();
    let module = cpu_module(&config)?;
    let data = FinePreprocessData {
        hw0_f: (4, 4),
        hw0_c: (2, 2),
        b_ids: Tensor::from_slice(&[0_i64, 0, 0]),
        i_ids: Tensor::from_slice(&[0_i64, 1, 3]),
        j_ids: Tensor::from_slice(&[1_i64, 2, 0]),
    };
    let fine_map0 = Tensor::randn([1, 128, 4, 4], (Kind::Float, Device::Cpu));
    let fine_map1 = Tensor::randn([1, 128, 4, 4], (Kind::Float, Device::Cpu));
    let coarse_tokens0 = Tensor::randn([1, 4, 256], (Kind::Float, Device::Cpu));
    let coarse_tokens1 = Tensor::randn([1, 4, 256], (Kind::Float, Device::Cpu));

    let (out0, out1) = module.forward(
        &fine_map0,
        &fine_map1,
        &coarse_tokens0,
        &coarse_tokens1,
        &data,
    )?;
    assert_eq!(out0.size(), vec![3, 25, 128]);
    assert_eq!(out1.size(), vec![3, 25, 128]);
    Ok(())
}

#[test]
fn stride_matches_kornia_height_axis_behavior() -> Result<(), LoftrError> {
    let data = FinePreprocessData {
        hw0_f: (8, 12),
        hw0_c: (4, 4),
        b_ids: Tensor::zeros([0], (Kind::Int64, Device::Cpu)),
        i_ids: Tensor::zeros([0], (Kind::Int64, Device::Cpu)),
        j_ids: Tensor::zeros([0], (Kind::Int64, Device::Cpu)),
    };
    assert_eq!(data.stride()?, 2);
    Ok(())
}