BgFilter

中文: README_ZH.md

Offline character matting for AI-generated images on a flat-colour background. The background colour is auto-detected from the image border (green, pastel, any flat colour); pass --screen-color to set it explicitly.

RGB input
  -> chroma bg-confidence (keyed to the auto-detected or given colour)
  -> [optional] semantic segmentation mask
  -> trimap -> ViTMatte -> alpha cleanup
  -> cross-model veto (second matting opinion on bg-hued residue)
  -> pymatting foreground -> despill -> RGBA PNG -> QA previews

Pipelines

Two pipelines, selected by segmentation.enabled in the config:

• Chroma-only (enabled: false) — Chroma + ViTMatte, no segmentation model. Lightest / fastest; leans entirely on the colour key for topology.
• Single segmenter (enabled: true, default) — one segmentation model drives the trimap topology (holes, hair), ViTMatte then refines the soft edges. Backend is switchable: birefnet (default, general salient objects — text, logos, photos; needs trust_remote_code) or anime-seg (ONNX, tuned for anime characters). Switch at runtime with --seg-backend anime-seg (it also selects the matching weights).

Both share pymatting foreground estimation and a colour de-spill. There is no green-contamination rescue / recolour layer — with clean source images it is unnecessary, so it was removed.

Both also run a cross-model veto by default: a second, trimap-free matting model (ZhengPeng7/BiRefNet_HR-matting) may only lower alpha, only on background-hued bright pixels the primary result is confident about — this clears colour-drifted background residue trapped between hair strands that the chroma key, the segmenter and ViTMatte all read as foreground. Costs one extra model download (~0.9 GB) and one inference pass per image; disable with --no-cross-check (see the cross_check config section, and docs/hair_gap_artifacts.md for the analysis behind it).

When the cross-check is on, that same HR-matting forward can optionally be reused as the segmentation mask (--cross-check-as-seg), skipping the primary seg model entirely (one less model to load, ~20 s faster per image on CPU, ~1 GB less VRAM on GPU). Pilot-validated (TestImage3 / FixImage1): trimap 99.8% identical, no structural change to fingers, hair wisps or thin lines. Off by default: the dedicated segmenter keeps its own forward and the veto stays an independent second signal. The reuse is a same-family swap, so it applies to the birefnet backend only: with --seg-backend anime-seg the anime segmenter always keeps its own forward.

The segmentation trimap defaults to directional mode (chroma + seg + a hue-direction split: it keeps a background-coloured garment such as a white shirt while dropping a background-hued residual such as blue trapped between hair strands). Switch with --trimap-mode seg (topology only, no hue split) or directional-hard-bg (aggressive — hard-removes background-hued pixels; can eat cool/shadowed white cloth).

Background colour

By default (screen_color: null) the background colour is auto-detected from the image border: the dominant flat colour of the border strip becomes the key colour. If the border is not one clean flat colour — a gradient, texture, or a subject filling the frame — detection fails with an error; pass --screen-color explicitly in that case.

To set it yourself, give a hex prior:

... --screen-color "#CFEFFF"

or screen_color: "#CFEFFF" in the config. Either way the chroma key scores pixels by perceptual (Lab/RGB) distance to the colour, and de-spill removes chroma along that colour's direction. (A supplied hex is refined against nearby border pixels; an auto-detected colour is used directly.)

Environment

Use the conda environment lightML.

conda activate lightML
pip install -r requirements.txt

If the shell is not activated, call the environment Python directly:

D:\MiniConda\envs\lightML\python.exe -m bgfilter.cli --help

Single Image

D:\MiniConda\envs\lightML\python.exe -m bgfilter.cli `
  --input Samples\TestImage.png `
  --output Outputs\TestImage_rgba.png `
  --debug-dir Outputs\TestImage_debug

No --config is needed — the built-in defaults are identical to configs/default.yaml. Pass --config configs\default.yaml only after you edit that file to tune the detailed parameters. Runtime choices stay on the command line: --device (default CPU; use --device cuda for GPU), --precision (default fp32; bf16 speeds up all three models and halves the matting model's activation memory with visually identical alpha — needs bf16-capable hardware, falls back to fp32 elsewhere; large inputs additionally get ViTMatte's global attention computed in query chunks by default — exact, bitwise-identical, caps the memory spike at ~4 GB instead of ~19 GB at 2048x2048, see model.attn_query_chunk), --seg-backend (default birefnet; anime-seg for anime characters), --screen-color (default: auto-detect the flat background), and --trimap-mode.

Batch

D:\MiniConda\envs\lightML\python.exe -m bgfilter.cli `
  --input-dir Samples `
  --output-dir Outputs `
  --debug-dir Outputs\debug

Performance (CPU reference numbers)

Measured on a Ryzen 9700X (Zen 5, native bf16), 32 GB RAM, with cross-check on, --cross-check-as-seg enabled and chunked global attention on. (Reuse is off by default on this branch; without it, add the primary BiRefNet@1024 forward — roughly +20 s/image at these sizes.)

input	precision	warm / image	peak memory
1024x1536	fp32	~45 s	—
1024x1536	bf16	~33 s	—
2048x2048	bf16	~53 s	11.4 GB (batch), 8.1 GB (single image)

The memory ceiling is the cross-check BiRefNet_HR forward (fixed input_size 2048 regardless of the image size) — genuine live activations of a full-resolution dense prediction net. Before chunked attention and the mimalloc fix, the same 2048x2048 bf16 run peaked at 25.2 GB; a fully un-optimized fp32 run would need an estimated 55-60 GB (ViTMatte's un-chunked N^2 attention alone ~39 GB). CPUs without native bf16 (e.g. Zen 2) auto-fall back to fp32 — on such machines prefer --no-cross-check if memory or latency is tight.

Chroma-alpha debug mode

--matting-method chroma skips ViTMatte and uses chroma confidence directly as the alpha seed. Useful for fast inspection of chroma confidence, trimap, and despill. (Distinct from the chroma-only pipeline above, which still runs ViTMatte.)

D:\MiniConda\envs\lightML\python.exe -m bgfilter.cli `
  --input-dir Samples `
  --output-dir Outputs\chroma `
  --debug-dir Outputs\chroma_debug `
  --matting-method chroma `
  --device cpu

Outputs

For each processed image, the CLI writes an RGBA PNG and optional debug files:

bg_confidence.png
trimap.png
seg_mask.png          # segmentation pipeline only
alpha.png
foreground_rgb.png    # despilled foreground colour
color_mask.png        # per-pixel despill weight
preview_black.png
preview_white.png
preview_gray.png
preview_red.png
preview_blue.png
qa_grid.png
metadata.json

HTTP service

An HTTP wrapper (app.py + bgfilter/service.py) exposes the pipeline as a long-running FastAPI service. Models load once and are reused across requests. For deploying to a Linux server (dependencies, weights, systemd, offline mode, auth), see DEPLOY.md.

python -m uvicorn app:app --host 127.0.0.1 --port 18083 --workers 1

Run a single worker (--workers 1) — each worker loads its own copy of the models.

Endpoints

• GET /healthz — liveness + config JSON, e.g. {"ok":true,"service":"bgfilter","version":"0.1.0","defaultSegModel":"birefnet","device":"cpu","crossCheck":true}. Always unauthenticated.
• POST /remove-background — multipart/form-data in, image/png (RGBA) out.

Authentication

Optional, opt-in via the BGFILTER_AUTH_TOKEN env var. When set, /remove-background requires header X-Genarrative-Image-Token: <token> (constant-time compare); a missing or wrong token returns 401. When unset the endpoint is open (a fronting proxy is expected to authenticate). /healthz is always open. See DEPLOY.md §10.

Request fields (POST /remove-background, multipart/form-data)

Field	Required	Values	Meaning
file	yes	image file	Source image (field name fixed as file).
screen_color	no	#RRGGBB, or omit/empty/auto	Background-colour prior; omit to auto-detect from the border.
seg_model	no	birefnet (default) / anime-seg	Segmentation backend.
cross_check	no	1/0 (also true/false/yes/no/on/off)	Per-request override of the cross-model veto; omit/empty = service default.

Response headers (on 200)

Header	Example	Meaning
X-BGFilter-Elapsed-Ms	2372	Server-side processing time (ms).
X-BGFilter-Seg-Model	birefnet	Segmenter actually used.
X-BGFilter-Screen-Color	#CFEFFF / auto	Background colour used.
X-BGFilter-Cross-Check	on / off	Effective cross-check state.

Status codes

Code	When
200	Success — body is the RGBA PNG.
400	Bad input: unreadable image, invalid screen_color/seg_model/cross_check, or auto-detect failed (pass screen_color).
401	Auth enabled and the token is missing or wrong.
413	Image exceeds BGFILTER_MAX_IMAGE_PIXELS (~4 MP by default).
500	Inference failed.
503	Service still loading models (not ready).

Examples

# default (birefnet + auto background colour); add the auth header if the service requires it
curl -sS -F "file=@input.png" \
  -H "X-Genarrative-Image-Token: <token>" \
  http://127.0.0.1:18083/remove-background -o output.png

# explicit background colour + anime segmenter, cross-check off for this request
curl -sS \
  -F "file=@input.png" \
  -F "screen_color=#CFEFFF" \
  -F "seg_model=anime-seg" \
  -F "cross_check=off" \
  http://127.0.0.1:18083/remove-background -o output.png

Config comes from BGFILTER_CONFIG (defaults to configs/default.yaml when present); BGFILTER_DEVICE overrides both model and segmentation device; BGFILTER_MAX_IMAGE_PIXELS caps input size (default ~4MP → 413); BGFILTER_PRELOAD=1 loads the default models at startup; BGFILTER_AUTH_TOKEN gates the endpoint (see Authentication above).

Like the CLI, the service runs the cross-model veto by default (config default: on); its HR-matting model is preloaded with the rest of the warm set. Set BGFILTER_CROSS_CHECK=0 to turn it off service-wide, or use the per-request cross_check form field to override either way; /healthz reports the default as crossCheck. With cross_check.reuse_as_seg: true in the config (off by default) the HR-matting forward replaces the primary seg forward instead of adding to it, and the then-redundant primary segmenter is skipped at startup.

Quality Check

The quality checker measures alpha validity and edge spill on semi-transparent edge pixels.

D:\MiniConda\envs\lightML\python.exe -m bgfilter.quality_cli `
  Outputs\TestImage_rgba.png `
  --max-edge-green-excess-p95 0.30

Run the bundled sample smoke check:

D:\MiniConda\envs\lightML\python.exe scripts\smoke_samples.py `
  --samples-dir Samples `
  --output-dir Outputs\smoke_samples `
  --config configs\default.yaml `
  --device cpu `
  --fallback-to-chroma-alpha `
  --max-edge-green-excess-p95 0.30

Notes

• Samples/ and Outputs/ are ignored by Git.
• ViTMatte and segmentation weights load from Hugging Face on first use. anime-seg (skytnt/anime-seg) is a plain ONNX download; birefnet (ZhengPeng7/BiRefNet) ships custom modelling code so it needs trust_remote_code=True plus timm / einops / kornia.
• Downloading the weights behind a firewall — the reliable combination is the hf-mirror.com mirror with the local proxy bypassed and Xet disabled:

  $env:HF_ENDPOINT      = "https://hf-mirror.com"  # domestic mirror
  $env:NO_PROXY         = "*"                       # bypass the proxy; the mirror is direct
  $env:HF_HUB_DISABLE_XET = "1"                     # these repos are Xet-backed; force classic HTTP
  

  Two gotchas this avoids: (1) mirror + an overseas proxy makes the mirror's resolve bounce back to huggingface.co, which recent huggingface_hub rejects with FileMetadataError; (2) with hf-xet installed the Xet download path fails instantly. Once the weights are cached, run offline with HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 (as the service does in production).
• Bundling weights in the project — instead of the HF cache, drop each model into a plain folder named after the repo basename under models/: models/vitmatte-base-composition-1k, models/BiRefNet, models/anime-seg, models/BiRefNet_HR-matting (cross-check second opinion, default on). The loader prefers a matching local folder and falls back to the HF repo id / cache when absent, so it is opt-in. Populate them with e.g. hf download ZhengPeng7/BiRefNet --local-dir models/BiRefNet. Override the base directory with BGFILTER_WEIGHTS_DIR. models/ is gitignored.
• Foreground colour estimation uses pymatting's estimate_foreground_ml to propagate clean foreground colour into semi-transparent edges before de-spill. Set foreground.method: unmix to fall back to the legacy heuristic.
• docs/green_screen_matting_workflow.md is the original phase-1 green-screen spec; this README reflects the current, generalised architecture.