Update README.md

README.md

@@ -25,9 +25,10 @@ Traditional face models fail where it matters most for AI art workflows:
 |-------------|-------------------|
 | 🎨 **Domain-locked** | Existing models excel at *either* anime *or* realistic—never both |
 | 🔞 **NSFW blindness** | Most models trained only on SFW data break on adult content |
+| 👁️‍🗨️ **Detail blindness** | Most models miss anime eyebrows, real eyelashes etc. |
 | 🎲 **Generation artifacts** | Standard datasets don't include diffusion model quirks and failures |
 
-**These models solve all three.**
+**These models solve all 4.**
 
 ---
 
@@ -35,7 +36,7 @@ Traditional face models fail where it matters most for AI art workflows:
 
 ### The Dataset Difference
 
-Built from **11,000+ manually annotated images** across the domains that actually matter for AI generation:
+Built from **14,000+ manually annotated images** across the domains that actually matter for AI generation:
 
 <table>
 <tr>
@@ -79,11 +80,10 @@ These models: Trained on what you actually generate
 
 ### Face Detection (YOLOv11-Small)
 
-**Purpose:** Primary face detection with high recall and very tight face boxes
+**Purpose:** Primary face detection with high recall
 
 **Training Approach:**
 - After every training run, I ran the model on a new mixed dataset, hardmining failures and improving the dataset until an acceptable performance was reached
-- Used offline custom augmentation on the initial set to complement light yolo training script augmentations
 - Trained at 640px resolution (inference should use same resolution)
 
 **Why YOLOv11-Small instead of nano?**  
@@ -92,19 +92,21 @@ More reliable detection across mixed realistic/anime domains with acceptable spe
 ---
 
 
-### Segmentation (GhostNetV2-100)
+### Segmentation (EfficientNet-v2)
 
-**Purpose:** Precise face mask generation including hair and complex boundaries
+**Purpose:** Precise face mask generation
 
 **Training Approach:**
-- [TRAINING DETAILS PLACEHOLDER]
-- Pixel-accurate mask annotations
-- [AUGMENTATION DETAILS PLACEHOLDER]
+- Dataset prepared using the Forbidden Vision yolo model at 512px resolution
+- Iterative hardmine training in multiple phases:
+-- started with 700 samples, trained
+-- run trained model on untrained set of images -> pick failures -> correct -> include in total set and retrain
+-- repeat process until no obvious failures exist -> final set 4k+ images 
 
 **Features:**
-- Includes hair, face, and neck regions
-- Handles complex occlusions (hands, objects, accessories)
-- Smooth mask edges for seamless blending
+- Detects and includes facial features other models ignore, like protruding anime eybrows, realistic eyelashes sticking out of the face etc.
+- Glasses and similar are treated as part of the face, even if sticking outside the face shape
+- NSFW friendly across both anime, realistic and 3d domains
 
 ---