lhk229
dfc85566c7
bf16 halves the matting model's activation memory -- its full-resolution attention is the pipeline's memory peak -- with visually identical alpha (measured: 0 px alpha deviation > 0.25 on samples; cross-check veto behaviour unchanged, region IoU 0.984). Guarded by a hardware check so it never lands on a slow emulation path: CUDA requires torch.cuda.is_bf16_supported(); CPU requires the same oneDNN native-bf16 gate PyTorch uses for matmul routing (AVX512-BF16/ AMX). Without support it warns and falls back to fp32 -- on a Zen2 EPYC the fallback kernels measured 17-370x slower than fp32, so silent bf16 there would be a performance landmine. Segmentation stays fp32: torchvision deform_conv2d (used by BiRefNet) has no bf16 CPU kernel, and the segmenter is not the memory peak. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
92 lines
3.5 KiB
YAML
92 lines
3.5 KiB
YAML
# Background-colour prior. null = auto-detect the flat background colour from the
|
|
# image border (raises if the border is not one clean flat colour). A hex string
|
|
# like "#CFEFFF" sets it explicitly. Also overridable via --screen-color.
|
|
screen_color: null
|
|
|
|
model:
|
|
model_name: hustvl/vitmatte-base-composition-1k
|
|
device: cpu
|
|
matting_method: vitmatte
|
|
fallback_to_chroma_alpha: false
|
|
# "bf16" halves ViTMatte activation memory (visually identical alpha); needs
|
|
# bf16-capable hardware (modern GPU / AVX512-BF16 CPU), else falls back to fp32.
|
|
precision: fp32
|
|
|
|
segmentation:
|
|
# enabled: true = single-segmenter pipeline (one seg model drives the trimap)
|
|
# enabled: false = no seg model, chroma-only pipeline (Chroma + ViTMatte)
|
|
enabled: true
|
|
backend: birefnet # single trimap segmenter; switchable at runtime: --seg-backend birefnet|anime-seg
|
|
model_name: ZhengPeng7/BiRefNet # matches backend; --seg-backend anime-seg auto-selects skytnt/anime-seg
|
|
device: cpu
|
|
input_size: 1024
|
|
|
|
trimap:
|
|
sure_bg_threshold: 0.92
|
|
sure_fg_threshold: 0.12
|
|
unknown_radius_ratio: 0.012
|
|
fg_safe_radius_ratio: 0.006
|
|
seg_core_threshold: 0.60
|
|
seg_loose_threshold: 0.08
|
|
# Trimap mode (segmentation pipeline only; also settable via --trimap-mode):
|
|
# directional = chroma + seg + hue split; background-hued residual
|
|
# (blue between hair) -> unknown, white shirt kept. (default)
|
|
# seg = original fuse_trimap (no hue split).
|
|
# directional-hard-bg = hard-remove background-hued pixels (aggressive; can eat
|
|
# cool/shadowed white cloth).
|
|
mode: directional
|
|
seg_low: 0.15
|
|
bg_hue_proj_min: 4.0
|
|
|
|
alpha_post:
|
|
chroma_suppress: true
|
|
chroma_suppress_bg_low: 0.35
|
|
chroma_suppress_bg_high: 0.80
|
|
chroma_suppress_strength: 1.0
|
|
# Matte-confidence gate: suppression fades where ViTMatte's raw alpha is high
|
|
# (ramp lo -> hi), so colour evidence only vetoes the matte where it is unsure.
|
|
# Set suppress_raw_lo: 1.0 to disable.
|
|
suppress_raw_lo: 0.85
|
|
suppress_raw_hi: 0.98
|
|
|
|
cross_check:
|
|
# Cross-model veto: a second, trimap-free matting model may only LOWER alpha
|
|
# (min-fusion), only on background-hued bright pixels the primary result is
|
|
# confident about — clears colour-drifted background residue between hair
|
|
# strands that chroma, segmentation and ViTMatte all read as foreground.
|
|
# Costs one extra model (~0.9 GB download) and one inference pass per image.
|
|
# Disable with enabled: false or --no-cross-check.
|
|
enabled: true
|
|
model_name: ZhengPeng7/BiRefNet_HR-matting
|
|
input_size: 2048
|
|
proj_min: 3.0 # bg-hue projection above which a pixel is suspect
|
|
l_min: 45.0 # Lab lightness below which a pixel is exempt (dark hair)
|
|
feather_sigma: 2.0 # zone-boundary feather, px
|
|
gate_lo: 0.70 # primary alpha below this -> fully exempt
|
|
gate_hi: 0.95 # primary alpha above this -> fully vetoable
|
|
|
|
foreground:
|
|
enabled: true
|
|
method: ml
|
|
fallback_to_unmix: true
|
|
ml_regularization: 0.00001
|
|
# Legacy "unmix" method params (used only when method: unmix).
|
|
edge_low: 0.005
|
|
edge_high: 0.995
|
|
min_unmix_alpha: 0.08
|
|
unmix_strength: 0.70
|
|
local_strength: 0.45
|
|
local_blur_radius: 11
|
|
local_alpha_threshold: 0.92
|
|
local_bg_confidence_max: 0.20
|
|
green_excess_margin: 0.015
|
|
bg_confidence_weight: 0.70
|
|
|
|
despill:
|
|
enabled: true
|
|
edge_low: 0.005
|
|
strength: 0.92
|
|
edge_expand_radius: 2
|
|
alpha_weight_floor: 0.35
|
|
color_margin_lab: 4.0 # Lab chroma margin along the background-colour direction
|