Add foreground color estimation pass

README.md

@@ -84,6 +84,10 @@ For each processed image, the CLI writes an RGBA PNG and optional debug files:
 bg_confidence.png
 trimap.png
 alpha.png
+foreground_mask.png
+foreground_unmix_mask.png
+foreground_local_mask.png
+foreground_rgb.png
 despill_mask.png
 preview_black.png
 preview_white.png
@@ -102,7 +106,7 @@ edge pixels.
 ```powershell
 D:\MiniConda\envs\lightML\python.exe -m bgfilter.quality_cli `
   Outputs\TestImage_rgba.png `
-  --max-edge-green-excess-p95 0.50
+  --max-edge-green-excess-p95 0.30
 ```
 
 Run the bundled sample smoke check:
@@ -114,13 +118,12 @@ D:\MiniConda\envs\lightML\python.exe scripts\smoke_samples.py `
   --config configs\default.yaml `
   --device cpu `
   --fallback-to-chroma-alpha `
-  --max-edge-green-excess-p95 0.50
+  --max-edge-green-excess-p95 0.30
 ```
 
 ## Current Notes
 
 - `Samples/` and `Outputs/` are ignored by Git.
 - `ViTMatte` model weights are loaded from Hugging Face on first use.
-- Current despill is edge-local and includes a lightweight foreground unmixing pass.
-  Strong green spill around hair may still need more advanced foreground color
-  estimation.
+- Foreground color estimation runs before final de-spill and writes debug masks
+  for unmixing and local foreground propagation.