Added weights file, configuration file, and a demo README.md

README.md

@@ -1,3 +1,78 @@
 ---
+language:
+  - en
+  - ru
+tags:
+  - computer-vision
+  - object-detection
+  - yolov8
+  - welding
+  - ndt
+  - defect-detection
 license: mit
+datasets:
+  - synthetic-welding-defects
+metrics:
+  - precision
+  - recall
+  - mAP
 ---
+
+# XVL: X-Ray Vision Lab - Welding Defect Detector
+
+YOLOv8-based model for automated detection of welding defects in X-ray images.
+
+## Model Details
+- **Architecture**: YOLOv8n (custom)
+- **Input Size**: 512x512
+- **Classes**: 5 defect types
+- **Training Data**: Synthetic X-ray images (8000+ samples)
+- **Validation Data**: Real industrial X-ray scans (200+)
+
+## Performance
+| Metric     | Value  | Epoch |
+|------------|--------|-------|
+| Precision  | 95.6%  | 37    |
+| Recall     | 88.9%  | 39    |
+| mAP@50     | 93.3%  | 39    |
+| mAP@50-95  | 78.8%  | 37    |
+
+## Usage
+
+### With PyTorch
+```python
+import torch
+from models.yolo_custom import load_model
+
+model = load_model(
+    weights="best.pt",
+    config="config.yaml"
+)
+With XVL Project
+bash
+git clone https://github.com/your-username/XVL.git
+python scripts/download_weights.py
+python src/run.py
+Defect Classes
+Cracks
+
+Porosity
+
+Lack of penetration
+
+Slag inclusions
+
+Undercut
+
+Training Configuration
+See config.yaml for full details.
+
+Citation
+If you use this model, please reference:
+
+@software{xvl2024,
+  title={XVL: X-Ray Vision Lab},
+  author={Your Name},
+  year={2024},
+  url={https://github.com/your-username/XVL}
+}

basic_config.yaml

@@ -0,0 +1,42 @@
+# configs.yaml
+training:
+  img_size: 512
+  model_name: "yolov8s.pt"
+  epochs: 150
+  batch_size: 48
+  learning_rate: 0.00005
+  num_folds: 5
+
+data:
+  samples_per_epoch: null
+  val_samples: null
+  class_weights: null
+  class_names:
+    - "incomplete_fusion"
+    - "crack"
+    - "single_pore"
+    - "cluster_pores"
+    - "empty"
+
+# Параметры обучения YOLO
+yolo_args:
+  save: true
+  save_period: 10
+  exist_ok: true
+  pretrained: true
+  optimizer: "AdamW"
+  weight_decay: 0.005
+  dropout: 0.1
+  mixup: 0.05
+  cutmix: 0.05
+  warmup_epochs: 20
+  cos_lr: true
+  label_smoothing: 0.1
+  patience: 30
+  verbose: true
+  # Другие параметры YOLO могут быть добавлены здесь
+
+paths:
+  project_dir: "models"
+  log_dir: "logs/training"
+  metrics_dir: "metrics"

best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72f3c76d2fe2c4fe6fbfa65af9de65da528bef33a46967e4f20b0c1cc78ccc84
+size 51997906