Spaces:
Sleeping
Sleeping
Download code/tests/test_dynamic_deutan_separation.py from acra-thesis/acra-backend: direct link, hf CLI and curl.
- Browser
- Download file 4.79 kB
-
https://huggingface.co/spaces/acra-thesis/acra-backend/resolve/main/code/tests/test_dynamic_deutan_separation.py
- Command line
-
hf download hf://spaces/acra-thesis/acra-backend/code/tests/test_dynamic_deutan_separation.py
-
curl -L -o test_dynamic_deutan_separation.py https://huggingface.co/spaces/acra-thesis/acra-backend/resolve/main/code/tests/test_dynamic_deutan_separation.py
4.79 kB
| """ | |
| test_dynamic_deutan_separation.py | |
| --------------------------------- | |
| Unit and regression tests for Tier 3 Dynamic Separation of Conflicting Colors | |
| in Deutans (severity-adaptive gaps, headroom allocation, and orthogonal chroma shifts). | |
| """ | |
| import numpy as np | |
| import pytest | |
| from pipeline.__init__ import _linear_to_lab_f32, _srgb_to_linear_f32, run_full_pipeline | |
| from pipeline.conflict import delta_e_ciede2000, detect_conflicts | |
| from pipeline.cvd_simulation import get_simulation_matrix | |
| from pipeline.reencoding import ( | |
| _effective_target_gap, | |
| _simulate_lab_center, | |
| reencode, | |
| enforce_rg_separation, | |
| ORTHOGONAL_DIRECTIONS, | |
| ) | |
| def _srgb_to_lab(rgb: list) -> np.ndarray: | |
| u = np.array([[rgb]], dtype=np.uint8) | |
| return _linear_to_lab_f32(_srgb_to_linear_f32(u))[0, 0] | |
| def test_effective_target_gap_scaling(): | |
| """Verify that effective target gap scales monotonically with severity.""" | |
| # Deutan scaling | |
| gap_mild = _effective_target_gap(0.4, "deutan", base_target=30.0) | |
| gap_mod = _effective_target_gap(0.7, "deutan", base_target=30.0) | |
| gap_full = _effective_target_gap(1.0, "deutan", base_target=30.0) | |
| assert 20.0 < gap_mild < 21.0, f"Expected gap ~20.4 at s=0.4, got {gap_mild}" | |
| assert gap_mild < gap_mod < gap_full, "Gap must increase monotonically with severity" | |
| assert np.isclose(gap_full, 30.0), f"Expected gap 30.0 at s=1.0, got {gap_full}" | |
| # Protan baseline is higher to account for L-cone loss | |
| gap_protan_mild = _effective_target_gap(0.4, "protan", base_target=30.0) | |
| assert gap_protan_mild > gap_mild, "Protan must maintain a higher baseline lightness gap" | |
| def test_headroom_aware_allocation_high_green(): | |
| """Green near the ceiling (L*=86) must not be pushed past l_max (90).""" | |
| # Create red at 45 and green at 86 (shortfall = 30 - (86 - 45) = -11 -> already separated, | |
| # let's set red at 70 and green at 86, gap = 16, shortfall = 14) | |
| centers = np.array([ | |
| [70.0, 35.0, 20.0], # red | |
| [86.0, -35.0, 25.0], # green near ceiling | |
| ]) | |
| out = enforce_rg_separation(centers, target_gap=30.0, l_min=2.0, l_max=90.0, severity=1.0, cvd_type="deutan") | |
| green_new = out[1, 0] | |
| red_new = out[0, 0] | |
| assert green_new <= 90.0, f"Green must not clip past l_max (90.0), got {green_new}" | |
| assert (green_new - red_new) >= 29.9, f"Total gap must reach target ~30, got {green_new - red_new}" | |
| assert red_new < 60.0, f"Red should absorb the majority of the downward shift, got {red_new}" | |
| def test_headroom_aware_allocation_dark_red(): | |
| """Red near the floor (L*=6) must not be pushed below l_min (2.0).""" | |
| centers = np.array([ | |
| [6.0, 15.0, 10.0], # dark red | |
| [20.0, -18.0, 15.0], # dark green | |
| ]) | |
| out = enforce_rg_separation(centers, target_gap=30.0, l_min=2.0, l_max=90.0, severity=1.0, cvd_type="deutan") | |
| red_new = out[0, 0] | |
| green_new = out[1, 0] | |
| assert red_new >= 2.0, f"Red must not drop below l_min (2.0), got {red_new}" | |
| assert (green_new - red_new) >= 29.9, f"Total gap must reach target ~30, got {green_new - red_new}" | |
| assert green_new > 30.0, f"Green should absorb the majority of the upward shift, got {green_new}" | |
| def test_deutan_orthogonal_chroma_shift(): | |
| """Verify that the deutan orthogonal vector produces significant separation.""" | |
| v_ortho = ORTHOGONAL_DIRECTIONS["deutan"] | |
| sim_m = get_simulation_matrix(1.0, "deutan") | |
| red = _srgb_to_lab([180, 50, 50]) | |
| green = _srgb_to_lab([50, 140, 50]) | |
| centers = np.array([red, green]) | |
| conflicts = detect_conflicts(centers, _simulate_lab_center(centers, sim_m), "deutan") | |
| assert len(conflicts) > 0, f"Expected conflict between red and green, got {conflicts}" | |
| reenc = reencode(centers, conflicts, 1.0, "deutan", allow_chroma_shift=True) | |
| sim_reenc = _simulate_lab_center(reenc, sim_m) | |
| de_post = delta_e_ciede2000(sim_reenc[0], sim_reenc[1]) | |
| assert de_post > 20.0, f"Re-encoded simulated difference must be suprathreshold (>20.0), got {de_post}" | |
| def test_mild_deutan_naturalness_advantage(): | |
| """Mild deuteranomaly (s=0.4) should achieve lower naturalness drift than severe (s=1.0).""" | |
| img = np.zeros((100, 100, 3), dtype=np.uint8) | |
| img[:50, :] = [200, 50, 50] | |
| img[50:, :] = [50, 200, 50] | |
| _, m_mild = run_full_pipeline(img, 0.4, "deutan", use_segmentation=False) | |
| _, m_severe = run_full_pipeline(img, 1.0, "deutan", use_segmentation=False) | |
| nat_mild = m_mild.get("naturalness_preservation", 999.0) | |
| nat_severe = m_severe.get("naturalness_preservation", 999.0) | |
| assert nat_mild < nat_severe, f"Mild deutan naturalness ({nat_mild:.2f}) must be better than severe ({nat_severe:.2f})" | |
| assert m_mild.get("conflict_resolution_rate", 0) == 1.0, "Mild deutan must achieve 100% resolution rate" | |