Spaces:
Sleeping
Sleeping
File size: 5,722 Bytes
78661be 8cbfe96 78661be | 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 | import numpy as np
import pytest
from pipeline.cvd_simulation import (
get_simulation_matrix,
simulate_cvd,
simulate_cvd_machado2009,
simulate_cvd_brettel1997,
simulate_cvd_vienot1999,
MACHADO_MATRICES,
)
def test_white_point_preservation():
"""White [1, 1, 1] must map to [1, 1, 1] under all severities and CVD types."""
for cvd in ("deutan", "protan"):
for sev in (0.0, 0.3, 0.5, 0.8, 1.0):
M = get_simulation_matrix(sev, cvd)
white_sim = np.array([1.0, 1.0, 1.0]) @ M.T
np.testing.assert_allclose(
white_sim, [1.0, 1.0, 1.0], atol=1e-3,
err_msg=f"White point not preserved for {cvd} at sev={sev}"
)
def test_machado2009_daltonlens_exact_parity():
"""Our vectorized Machado 2009 implementation must match DaltonLens reference perfectly."""
try:
import daltonlens.simulate as dl_sim
except ImportError:
pytest.skip("daltonlens library not available")
ref_sim = dl_sim.Simulator_Machado2009()
np.random.seed(42)
test_img = np.random.rand(16, 16, 3).astype(np.float32)
for cvd, dl_deficiency in [("deutan", dl_sim.Deficiency.DEUTAN), ("protan", dl_sim.Deficiency.PROTAN)]:
for sev in [0.0, 0.2, 0.5, 0.8, 1.0]:
our_out = simulate_cvd_machado2009(test_img, sev, cvd)
ref_out = ref_sim._simulate_cvd_linear_rgb(test_img, dl_deficiency, sev)
# Check exact match (< 1e-6)
max_diff = np.max(np.abs(our_out - ref_out))
assert max_diff < 1e-6, f"Discrepancy with DaltonLens Machado for {cvd} at s={sev}: {max_diff}"
def test_brettel1997_white_point_preservation():
"""Brettel 1997 must preserve white [1, 1, 1] across all severities."""
white = np.array([[[1.0, 1.0, 1.0]]], dtype=np.float32)
for cvd in ("deutan", "protan"):
for sev in (0.0, 0.3, 0.7, 1.0):
out = simulate_cvd_brettel1997(white, sev, cvd)
np.testing.assert_allclose(
out[0, 0], [1.0, 1.0, 1.0], atol=1e-3,
err_msg=f"Brettel white point drift for {cvd} at sev={sev}"
)
def test_brettel1997_daltonlens_exact_parity():
"""Our vectorized Brettel 1997 implementation must match DaltonLens reference perfectly."""
try:
import daltonlens.simulate as dl_sim
except ImportError:
pytest.skip("daltonlens library not available")
ref_sim = dl_sim.Simulator_Brettel1997()
np.random.seed(123)
test_img = np.random.rand(16, 16, 3).astype(np.float32)
for cvd, dl_deficiency in [("deutan", dl_sim.Deficiency.DEUTAN), ("protan", dl_sim.Deficiency.PROTAN)]:
for sev in [0.5, 1.0]:
our_out = simulate_cvd_brettel1997(test_img, sev, cvd)
ref_out = ref_sim._simulate_cvd_linear_rgb(test_img, dl_deficiency, sev)
# Both unclipped or clipped should match to within float precision (< 1e-5)
max_diff = np.max(np.abs(our_out - ref_out))
assert max_diff < 1e-5, f"Discrepancy with DaltonLens reference for {cvd} at s={sev}: {max_diff}"
def test_brettel1997_confusion_line_collapse():
"""Under deuteranopia, two colors along a deutan confusion line must collapse to the same simulated color."""
lms1 = np.array([0.2, 0.1, 0.01])
lms2 = np.array([0.2, 0.3, 0.01])
from pipeline.cvd_simulation import LINEAR_RGB_FROM_LMS
rgb1 = (lms1 @ LINEAR_RGB_FROM_LMS.T).reshape(1, 1, 3).astype(np.float32)
rgb2 = (lms2 @ LINEAR_RGB_FROM_LMS.T).reshape(1, 1, 3).astype(np.float32)
sim1 = simulate_cvd_brettel1997(rgb1, 1.0, "deutan")
sim2 = simulate_cvd_brettel1997(rgb2, 1.0, "deutan")
np.testing.assert_allclose(sim1, sim2, atol=1e-5)
def test_vienot1999_red_green_collapse():
"""Under Viénot 1999 full dichromacy, Row 0 and Row 1 are identical."""
for cvd in ("deutan", "protan"):
M = get_simulation_matrix(1.0, cvd, model="vienot1999")
np.testing.assert_allclose(M[0], M[1], atol=1e-4)
def test_severity_interpolation():
"""Severity 0.0 must be identity; intermediate values match DaltonLens Machado tables."""
for cvd in ("deutan", "protan"):
M_0 = get_simulation_matrix(0.0, cvd)
np.testing.assert_allclose(M_0, np.eye(3), atol=1e-6)
M_half = get_simulation_matrix(0.5, cvd)
np.testing.assert_allclose(M_half, MACHADO_MATRICES[cvd][5], atol=1e-6)
M_full = get_simulation_matrix(1.0, cvd)
np.testing.assert_allclose(M_full, MACHADO_MATRICES[cvd][10], atol=1e-6)
# A severity BETWEEN two table steps. 0.0, 0.5 and 1.0 all land on a
# multiple of 0.1, where the old divide-first alpha happened to be
# correct, so they could never catch it. 0.35 sits halfway between
# table rows 3 and 4 and fails outright if alpha is wrong.
M_035 = get_simulation_matrix(0.35, cvd)
expected = 0.5 * MACHADO_MATRICES[cvd][3] + 0.5 * MACHADO_MATRICES[cvd][4]
np.testing.assert_allclose(M_035, expected, atol=1e-6)
# Brettel model still linearly interpolates
M_brettel_half = get_simulation_matrix(0.5, cvd, model="brettel1997")
M_brettel_full = get_simulation_matrix(1.0, cvd, model="brettel1997")
np.testing.assert_allclose(M_brettel_half, 0.5 * np.eye(3) + 0.5 * M_brettel_full, atol=1e-6)
def test_simulate_cvd_output_format():
"""simulate_cvd should return sRGB values in [0, 1]."""
test_img = np.array([[[0.5, 0.2, 0.8]]], dtype=np.float32)
srgb_out = simulate_cvd(test_img, 0.8, "deutan")
assert srgb_out.shape == test_img.shape
assert srgb_out.dtype == np.float32
assert 0.0 <= srgb_out.min() and srgb_out.max() <= 1.0
|