ringgrid 0.5.6

Pure-Rust detector for coded ring calibration 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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218

use std::collections::{HashMap, HashSet};

use crate::DetectedMarker;

fn sort_by_confidence(markers: Vec<DetectedMarker>) -> Vec<DetectedMarker> {
    let mut markers = markers;
    markers.sort_by(|a, b| b.confidence.partial_cmp(&a.confidence).unwrap());
    markers
}

fn dedup_by_proximity(markers: Vec<DetectedMarker>, radius: f64) -> Vec<DetectedMarker> {
    let mut keep = vec![true; markers.len()];
    let r2 = radius * radius;

    for i in 0..markers.len() {
        if !keep[i] {
            continue;
        }

        for j in (i + 1)..markers.len() {
            if !keep[j] {
                continue;
            }
            let dx = markers[i].center[0] - markers[j].center[0];
            let dy = markers[i].center[1] - markers[j].center[1];
            if dx * dx + dy * dy < r2 {
                keep[j] = false;
            }
        }
    }

    markers
        .into_iter()
        .enumerate()
        .filter_map(|(index, marker)| keep[index].then_some(marker))
        .collect()
}

fn dedup_by_id_core(markers: Vec<DetectedMarker>) -> Vec<DetectedMarker> {
    let mut best_idx: HashMap<usize, usize> = HashMap::new();

    for (index, marker) in markers.iter().enumerate() {
        if let Some(id) = marker.id {
            match best_idx.get(&id) {
                Some(&prev) if markers[prev].confidence >= marker.confidence => {}
                _ => {
                    best_idx.insert(id, index);
                }
            }
        }
    }

    let keep_set: HashSet<usize> = best_idx.values().copied().collect();
    markers
        .into_iter()
        .enumerate()
        .filter_map(|(index, marker)| {
            let keep = marker.id.is_none() || keep_set.contains(&index);
            keep.then_some(marker)
        })
        .collect()
}

/// Remove duplicate detections: keep the highest-confidence marker within `radius`.
pub fn dedup_markers(markers: Vec<DetectedMarker>, radius: f64) -> Vec<DetectedMarker> {
    let markers = sort_by_confidence(markers);
    dedup_by_proximity(markers, radius)
}

/// Dedup by ID: if the same decoded ID appears multiple times, keep the
/// one with the highest confidence.
pub fn dedup_by_id(markers: &mut Vec<DetectedMarker>) {
    let input = std::mem::take(markers);
    *markers = dedup_by_id_core(input);
}

// ---------------------------------------------------------------------------
// Multi-scale merge
// ---------------------------------------------------------------------------

fn outer_radius(m: &DetectedMarker) -> f64 {
    m.ellipse_outer
        .as_ref()
        .map(|e| e.mean_axis())
        .unwrap_or(0.0)
}

/// Compute a size-consistency score in `[0, 1]` for each marker.
///
/// The score measures how well the marker's outer radius matches the median
/// outer radius of its `k` nearest spatial neighbors. A score of 1.0 means
/// the radius equals the neighborhood median; lower values indicate
/// outliers (possible wrong-tier detections).
fn size_consistency_scores(markers: &[DetectedMarker], k_neighbors: usize) -> Vec<f64> {
    let n = markers.len();
    let k = k_neighbors.min(n.saturating_sub(1));
    let radii: Vec<f64> = markers.iter().map(outer_radius).collect();
    let mut scores = vec![1.0f64; n];

    if k == 0 || n <= 1 {
        return scores;
    }

    for i in 0..n {
        let ci = markers[i].center;

        // k nearest neighbors by squared Euclidean distance.
        let mut dists: Vec<(f64, usize)> = (0..n)
            .filter(|&j| j != i)
            .map(|j| {
                let cj = markers[j].center;
                let dx = ci[0] - cj[0];
                let dy = ci[1] - cj[1];
                (dx * dx + dy * dy, j)
            })
            .collect();
        dists.sort_by(|a, b| a.0.partial_cmp(&b.0).unwrap());

        let neighbor_radii: Vec<f64> = dists[..k.min(dists.len())]
            .iter()
            .map(|(_, j)| radii[*j])
            .collect();

        if neighbor_radii.len() < 3 {
            // Too few neighbors for a reliable median; leave score neutral.
            continue;
        }

        let mut sorted_nr = neighbor_radii.clone();
        sorted_nr.sort_by(|a, b| a.partial_cmp(b).unwrap());
        let mid = sorted_nr.len() / 2;
        let median_r = if sorted_nr.len().is_multiple_of(2) {
            (sorted_nr[mid - 1] + sorted_nr[mid]) * 0.5
        } else {
            sorted_nr[mid]
        };

        if median_r <= 0.0 {
            continue;
        }

        let r_i = radii[i];
        scores[i] = (1.0 - (r_i - median_r).abs() / median_r).max(0.0);
    }

    scores
}

/// Merge markers from multiple scale tiers using size-consistency-aware dedup.
///
/// For each pair of markers within `dedup_radius` pixels of each other
/// (spatial duplicates from overlapping tiers), keeps the one whose outer
/// radius is most consistent with the median outer radius of its `k_neighbors`
/// nearest neighbors. Ties are broken by decode confidence.
///
/// After spatial dedup, per-ID dedup keeps the highest-scoring match per
/// decoded ID.
///
/// `k_neighbors = 6` matches the hex-lattice valence and mirrors the
/// convention in [`InnerAsOuterRecoveryConfig`](super::config::InnerAsOuterRecoveryConfig).
pub(crate) fn merge_multiscale_markers(
    all_markers: Vec<DetectedMarker>,
    dedup_radius: f64,
    k_neighbors: usize,
) -> Vec<DetectedMarker> {
    if all_markers.is_empty() {
        return all_markers;
    }

    let n = all_markers.len();
    let scores = size_consistency_scores(&all_markers, k_neighbors);

    // Sort indices: highest size-consistency score first; confidence as tiebreak.
    let mut order: Vec<usize> = (0..n).collect();
    order.sort_by(|&i, &j| {
        scores[j]
            .partial_cmp(&scores[i])
            .unwrap_or(std::cmp::Ordering::Equal)
            .then_with(|| {
                all_markers[j]
                    .confidence
                    .partial_cmp(&all_markers[i].confidence)
                    .unwrap_or(std::cmp::Ordering::Equal)
            })
    });

    // NMS: iterate in score order; suppress lower-ranked markers within radius.
    let r2 = dedup_radius * dedup_radius;
    let mut keep = vec![true; n];

    for pos in 0..order.len() {
        let i = order[pos];
        if !keep[i] {
            continue;
        }
        let ci = all_markers[i].center;
        for &j in &order[pos + 1..] {
            if !keep[j] {
                continue;
            }
            let cj = all_markers[j].center;
            let dx = ci[0] - cj[0];
            let dy = ci[1] - cj[1];
            if dx * dx + dy * dy < r2 {
                keep[j] = false;
            }
        }
    }

    let surviving: Vec<DetectedMarker> = all_markers
        .into_iter()
        .enumerate()
        .filter_map(|(i, m)| keep[i].then_some(m))
        .collect();

    // Per-ID dedup: keep the best match per decoded ID (by confidence).
    dedup_by_id_core(surviving)
}