Update README.md

README.md

@@ -7,22 +7,17 @@ base_model:
 - Ultralytics/YOLO11
 pipeline_tag: image-segmentation
 ---
-# Car Damage Segmentation YOLOv11 Model
+# Car Damage Segmentation Model (YOLOv11-seg)
 
-## Model Overview
-This model is a fine-tuned YOLOv11-seg specifically optimized for vehicle damage segmentation. It can accurately detect and segment various types of vehicle damage using the CarDD dataset.
+This model detects and segments six types of vehicle damage: cracks, dents, glass shatter, broken lamps, scratches, and flat tires. Fine-tuned on the CarDD dataset, it provides precise damage localization and segmentation to support vehicle inspection and repair estimation.
 
 ## Model Details
-- **Base Architecture**: YOLOv11-seg
-- **Training Dataset**: CarDD dataset (Car Damage Detection)
-- **Input Resolution**: 1280×1280
-- **Damage Classes**:
-  - crack
-  - dent
-  - glass shatter
-  - lamp broken
-  - scratch
-  - tire flat
+
+- **Architecture**: YOLOv11-seg
+- **Base Model**: yolo11x-seg.pt
+- **Task**: Instance Segmentation
+- **Domain**: Automotive Damage Assessment
+- **Framework**: Ultralytics YOLOv11
 
 ## Performance Metrics
 
@@ -32,16 +27,6 @@ This model is a fine-tuned YOLOv11-seg specifically optimized for vehicle damage
 | Box  | 0.753     | 0.689  | 0.734 | 0.513    |
 | Mask | 0.762     | 0.692  | 0.735 | 0.503    |
 
-### Class-Specific Performance (Box Detection)
-| Class         | Precision | Recall | mAP50 | mAP50-95 |
-|---------------|-----------|--------|-------|----------|
-| crack         | 0.649     | 0.362  | 0.452 | 0.249    |
-| dent          | 0.607     | 0.524  | 0.565 | 0.324    |
-| glass shatter | 0.879     | 1.000  | 0.986 | 0.656    |
-| lamp broken   | 0.749     | 0.816  | 0.838 | 0.639    |
-| scratch       | 0.673     | 0.563  | 0.595 | 0.327    |
-| tire flat     | 0.958     | 0.871  | 0.971 | 0.880    |
-
 ### Class-Specific Performance (Mask Segmentation)
 | Class         | Precision | Recall | mAP50 | mAP50-95 |
 |---------------|-----------|--------|-------|----------|
@@ -52,67 +37,92 @@ This model is a fine-tuned YOLOv11-seg specifically optimized for vehicle damage
 | scratch       | 0.684     | 0.560  | 0.589 | 0.284    |
 | tire flat     | 0.958     | 0.869  | 0.971 | 0.891    |
 
-### Dataset Stats
-- **Images**: 571
-- **Instances**: 1247
-- **Instance Distribution**:
-  - crack: 152
-  - dent: 366
-  - glass shatter: 91
-  - lamp broken: 103
-  - scratch: 482
-  - tire flat: 53
-
-## Training Configuration
-```yaml
-batch: 24
-imgsz: 1280
-epochs: 300
-patience: 15
-workers: 16
-device: cuda
-cache: True
-amp: True
-overlap_mask: True
-warmup_epochs: 3
-mosaic: 1.0
-mixup: 0.1
-copy_paste: 0.1
-```
+## Training Recipe
 
-## Usage
+```python
+from ultralytics import YOLO
 
-### Installation
-```bash
-pip install ultralytics
+model = YOLO('yolo11x-seg.pt')
+
+model.train(
+    data='path/to/dataset.yaml',
+    epochs=300,
+    batch=16,
+    imgsz=1024,
+    dropout=0.1,
+    multi_scale=True,
+    val=True,
+    patience=25,
+    optimizer='AdamW',
+    cls=0.3,              # Reduced to prioritize boundary precision over classification
+    dfl=1.7,              # Increased for better damage boundary delineation
+    lr0=0.0001,
+    lrf=0.01,
+    warmup_epochs=10,     # Extended warmup for complex damage patterns
+    mosaic=1.0,
+    mixup=0.1,
+    copy_paste=0.1,
+    hsv_h=0.015,
+    hsv_s=0.7,
+    hsv_v=0.4,
+    cos_lr=True,
+    amp=True,
+    overlap_mask=True     # Better segmentation for irregular damage shapes
+)
 ```
 
-### Inference with Python
-```python
-from ultralytics import YOLO
+### Training Recipe for Car Damage Segmentation
 
-# Load the model
-model = YOLO('best.pt')
+- `imgsz=1024`: Higher resolution critical for detecting small damage features like fine cracks and scratches
 
-# Run inference on an image
-results = model('path/to/car/image.jpg')
+- `optimizer='AdamW'`: Provides better convergence for fine-grained segmentation tasks compared to SGD
 
-# Process results
-for result in results:
-    boxes = result.boxes  # Bounding boxes
-    masks = result.masks  # Segmentation masks
-```
+- `dropout=0.1`: Light regularization prevents overfitting while preserving ability to detect subtle damage features
+
+- `multi_scale=True`: Enhances model robustness to different damage sizes and viewing distances
+
+- `warmup_epochs=10`: Extended warmup period stabilizes early training on complex damage patterns
+
+- `overlap_mask=True`: Improves segmentation quality for irregular damage shapes with complex boundaries
+
+
+##### `cls=0.3`
 
-## Model Strengths
-- Excellent performance on glass shatter (98.6% mAP50) and tire flat (97.1% mAP50)
-- Good balance between precision and recall for most damage types
-- High-quality segmentation masks, particularly for well-defined damage types
+• Reduced classification weight prioritizes boundary precision over class distinction
 
-## Model Limitations
-- Lower recall for crack detection (36.2%)
-- Performance varies across damage types (mAP50 ranges from 45.2% for cracks to 98.6% for glass shatter)
-- Small damages might be missed, particularly fine cracks
-- Works best with clear, well-lit images of vehicles
+• Similar visual appearance between damage types (scratches vs. cracks)
+
+• Prevents overfitting to specific damage classes
+
+• Improves localization and segmentation accuracy
+
+##### `dfl=1.7`
+
+• Increased DFL for better segmentation of irregular damage boundaries
+
+• Addresses complex boundaries of car damage shapes
+
+• Improves bounding box precision
+
+• Effective for elongated damages like scratches
+
+
+## Dataset
+
+The model was trained on the CarDD (Car Damage Detection) dataset with:
+- 571 images
+- 1247 total damage instances
+- Class distribution: scratches (482), dents (366), cracks (152), broken lamps (103), glass shatter (91), and flat tires (53)
+
+## Usage
+
+## Limitations
+
+- Lower recall for crack detection (36.8%)
+- Performance varies across damage types
+- Small or fine damages may be missed
+- Best results with well-lit, clear images of vehicles
 
 ## License
-This model is available under the AGPL-3.0 License.
+
+Since this model was trained using Ultralytics, it falls under the AGPL-3.0 License.