File size: 6,713 Bytes
beb4f15
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Shared numeric and image helper utilities used across Fiberpose modules."""

import colorsys
import numpy as np
from numba import njit



def label_to_rgb(label: int) -> tuple[int, int, int]:
    """Deterministically map a positive label ID to a vivid RGB color."""
    if label <= 0:
        return (0, 0, 0)
    hue = (label * 0.6180339887498949) % 1.0  # golden ratio multiplier for good spacing
    sat = 0.55 + 0.35 * (((label * 0.2171) % 1.0))
    val = 0.85
    r, g, b = colorsys.hsv_to_rgb(hue, sat, val)
    return int(round(r * 255)), int(round(g * 255)), int(round(b * 255))


def labels_to_color(labels: np.ndarray | None) -> np.ndarray | None:
    """Render a label-id map into an RGB visualization image."""
    if labels is None:
        return None
    labels_i64 = np.asarray(labels, dtype=np.int64)
    if labels_i64.ndim != 2:
        return None
    if labels_i64.size == 0:
        return np.zeros(labels_i64.shape + (3,), dtype=np.uint8)
    # One color per distinct id (label_to_rgb maps ids <= 0 to black), then a single gather.
    ids, inverse = np.unique(labels_i64, return_inverse=True)
    palette = np.array([label_to_rgb(int(v)) for v in ids], dtype=np.uint8).reshape(-1, 3)
    return palette[inverse.reshape(labels_i64.shape)]

# ------------- image helpers -------------
def to_uint8_for_view(img: np.ndarray | None) -> np.ndarray | None:
    if img is None:
        return None
    arr = img.astype(np.float32, copy=False)
    if np.isnan(arr).any() or np.isinf(arr).any():
        arr = np.nan_to_num(arr, nan=0.0, posinf=0.0, neginf=0.0)
    amin, amax = float(arr.min()), float(arr.max())
    if amax <= amin:
        return np.zeros(arr.shape, dtype=np.uint8)
    arr = (arr - amin) / (amax - amin)
    return (arr * 255.0).clip(0, 255).astype(np.uint8)


def ensure_rgb_u8(img: np.ndarray | None) -> np.ndarray | None:
    if img is None:
        return None
    if img.ndim == 2:
        u8 = to_uint8_for_view(img)
        return np.stack([u8, u8, u8], axis=-1)
    if img.ndim == 3:
        if img.shape[2] == 1:
            ch = to_uint8_for_view(img[..., 0])
            return np.stack([ch, ch, ch], axis=-1)
        if img.dtype != np.uint8:
            return to_uint8_for_view(img)[..., :3]
        return img[..., :3]
    return ensure_rgb_u8(to_uint8_for_view(img))

# ------------- core math helpers -------------
@njit(cache=True)
def _normalize(v, eps=1e-12):
    n = np.sqrt(np.sum(v * v))
    return v / (n + eps)

@njit(cache=True)
def _bilinear(arr, y, x):
    """Bilinear sample arr at (y, x) with clamped borders. arr is HxW."""
    h, w = arr.shape
    y0 = int(np.floor(y))
    x0 = int(np.floor(x))
    y1 = y0 + 1
    x1 = x0 + 1

    y0c = 0 if y0 < 0 else (h - 1 if y0 >= h else y0)
    y1c = 0 if y1 < 0 else (h - 1 if y1 >= h else y1)
    x0c = 0 if x0 < 0 else (w - 1 if x0 >= w else x0)
    x1c = 0 if x1 < 0 else (w - 1 if x1 >= w else x1)

    wy = y - y0
    wx = x - x0
    a = arr[y0c, x0c]
    b = arr[y0c, x1c]
    c = arr[y1c, x0c]
    d = arr[y1c, x1c]
    return (1-wy) * ((1-wx)*a + wx*b) + wy * ((1-wx)*c + wx*d)

@njit(cache=True)
def _sample_vec2(gx, gy, y, x):
    """Sample gradient vector [dy, dx] at (y,x)."""
    return np.array([_bilinear(gy, y, x), _bilinear(gx, y, x)])

@njit(cache=True)
def _tangent2d(centerline, k):
    """Unit tangent at index k using central/one-sided differences. centerline is Nx2 in (y,x)."""
    N = len(centerline)
    if k == 0:
        t = centerline[1] - centerline[0]
    elif k == N-1:
        t = centerline[-1] - centerline[-2]
    else:
        t = centerline[k+1] - centerline[k-1]
    return _normalize(t)

@njit(cache=True)
def _rot90_ccw(v):
    """Rotate (y,x) by +90° in image coordinates."""
    return np.array([-v[1], v[0]])

# ------------- normals from DT -------------
@njit(cache=True)
def normals_from_dt_2d(centerline, dt, gx, gy, alpha=0.5, iters=3, tau_g=1e-3, eps=1e-8):
    """
    centerline: (N,2) float array of (y,x) points (same pixel coords as dt)
    dt: HxW float distance transform
    gx, gy: HxW gradients of dt (e.g., gx, gy = np.gradient(dt))
    alpha: probe step as fraction of local radius (h = alpha * r)
    iters: fixed-point iterations per point
    tau_g: threshold on |(g+ - g-)| below which we fall back to n0/previous
    returns: (N,2) unit normals (y,x) with continuity enforced
    """
    N = len(centerline)
    normals = np.zeros((N, 2), dtype=np.float64)
    n_iter = max(1, iters)

    for k in range(N):
        y, x = centerline[k]
        r = _bilinear(dt, y, x)
        h = alpha * max(r, 1e-6)  # always probe a tiny bit, even if r~0

        # initial guess n0 from tangent (rotate by 90°)
        t = _tangent2d(centerline, k)
        n = _rot90_ccw(t)

        # prefer previous orientation if available (continuity)
        if k > 0 and np.dot(n, normals[k-1]) < 0:
            n = -n

        # fixed-point refinement
        for _ in range(n_iter):
            yp, xp = y + h * n[0], x + h * n[1]
            ym, xm = y - h * n[0], x - h * n[1]
            gp = _sample_vec2(gx, gy, yp, xp)   # ∇DT at +h along n
            gm = _sample_vec2(gx, gy, ym, xm)   # ∇DT at -h along n
            d = 0.5 * (gp - gm)                 # averaged direction (normal candidate)
            mag = np.linalg.norm(d)
            if mag < tau_g:
                break
            n_new = d / (mag + eps)
            # keep orientation continuous along the curve
            if k > 0 and np.dot(n_new, normals[k-1]) < 0:
                n_new = -n_new
            n = n_new

        normals[k] = _normalize(n)

    return normals

# ------------- geometry helpers -------------
def resample_polyline_xy(points_xy: np.ndarray, step: float) -> np.ndarray:
    """Resample an open polyline (N,2) in XY to ~equal arc-length spacing of 'step'."""
    pts = np.asarray(points_xy, dtype=np.float64).reshape(-1, 2)
    if len(pts) < 2:
        return pts.copy()

    seg = pts[1:] - pts[:-1]
    seg_len = np.linalg.norm(seg, axis=1)
    cum = np.concatenate([[0.0], np.cumsum(seg_len)])
    L = cum[-1]
    if L <= 1e-6:
        return pts[:1].copy()

    s_targets = np.arange(0.0, L + 1e-9, float(step))
    if s_targets[-1] < L:
        s_targets = np.append(s_targets, L)

    out = []
    k = 0
    for s in s_targets:
        while k + 1 < len(cum) and cum[k+1] < s:
            k += 1
        if k + 1 >= len(cum):
            out.append(pts[-1])
        else:
            ds = cum[k+1] - cum[k]
            if ds <= 1e-12:
                out.append(pts[k].copy())
            else:
                a = (s - cum[k]) / ds
                out.append(pts[k] * (1 - a) + pts[k+1] * a)
    return np.asarray(out, dtype=np.float64)