Files
BGfilter/bgfilter/settings.py
T
lhk229 9b7a192599 Use pymatting ML foreground estimation to remove edge spill
After the full-foreground despill (5ddf0cd) the residual green was confined to
semi-transparent hair (alpha < 0.5): 85% of the worst pixels had alpha < 0.3,
where a pixel-wise unmix (divide by small alpha) is too noisy to trust.

Replace the hand-rolled unmix/local-blur estimator with pymatting's
estimate_foreground_ml (Germer et al. multi-level closed form), which propagates
reliable foreground colour from high-alpha neighbours into the fringe and
estimates the background, so green spill is unmixed rather than clamped. The
despill pass stays as a light cleanup on top.

- foreground.method selects "ml" (default) or "unmix" (legacy heuristic kept as
  a fallback when pymatting is unavailable, matching the chroma fallback idiom).
- ForegroundEstimate now exposes rgb (F), background (B) and a correction map;
  debug outputs become foreground_rgb / foreground_background / foreground_correction.

Controlled comparison (same ViTMatte alpha, only foreground method changed),
edge_green_excess:
  TestImage  p95 0.176->0.031, mean 0.061->0.005
  TestImage2 p95 0.165->0.004, mean 0.052->0.002

Adds pymatting (pulls in numba) to requirements.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-30 17:33:35 +08:00

83 lines
2.3 KiB
Python

from __future__ import annotations
from dataclasses import dataclass
@dataclass(frozen=True)
class ChromaSettings:
green_margin: float = 0.08
min_green_value: float = 0.22
min_saturation: float = 0.20
hue_low: float = 35.0
hue_high: float = 95.0
border_ratio: float = 0.04
min_samples: int = 2048
hue_sigma_min: float = 7.0
lab_sigma_min: float = 10.0
rgb_sigma_min: float = 0.08
@dataclass(frozen=True)
class TrimapSettings:
sure_bg_threshold: float = 0.92
sure_fg_threshold: float = 0.12
unknown_radius_ratio: float = 0.012
fg_safe_radius_ratio: float = 0.006
min_unknown_radius: int = 4
min_fg_safe_radius: int = 2
@dataclass(frozen=True)
class AlphaPostSettings:
min_component_area_ratio: float = 0.00001
fill_hole_area_ratio: float = 0.00002
alpha_floor: float = 0.002
alpha_ceil: float = 0.998
@dataclass(frozen=True)
class ForegroundSettings:
enabled: bool = True
method: str = "ml" # "ml" (pymatting) or "unmix" (legacy heuristic)
fallback_to_unmix: bool = True
ml_regularization: float = 1e-5
# Legacy "unmix" method parameters (used only when method == "unmix").
edge_low: float = 0.005
edge_high: float = 0.995
min_unmix_alpha: float = 0.08
unmix_strength: float = 0.70
local_strength: float = 0.45
local_blur_radius: int = 11
local_alpha_threshold: float = 0.92
local_bg_confidence_max: float = 0.20
green_excess_margin: float = 0.015
bg_confidence_weight: float = 0.70
@dataclass(frozen=True)
class DespillSettings:
enabled: bool = True
edge_low: float = 0.005
strength: float = 0.92
green_excess_margin: float = 0.015
edge_expand_radius: int = 2
alpha_weight_floor: float = 0.35
@dataclass(frozen=True)
class ModelSettings:
model_name: str = "hustvl/vitmatte-small-composition-1k"
device: str = "cuda"
matting_method: str = "vitmatte"
fallback_to_chroma_alpha: bool = False
@dataclass(frozen=True)
class PipelineSettings:
chroma: ChromaSettings = ChromaSettings()
trimap: TrimapSettings = TrimapSettings()
alpha_post: AlphaPostSettings = AlphaPostSettings()
foreground: ForegroundSettings = ForegroundSettings()
despill: DespillSettings = DespillSettings()
model: ModelSettings = ModelSettings()