restructure project

docs/training-flow.mmd

@@ -0,0 +1,54 @@
+flowchart TD
+    A[CLI Entry Point\nsrc/cli/train.py] --> B[Parse Arguments\n--model, --epochs, --batch, --imgsz, etc.]
+    B --> C[Connect PostgreSQL\nDB_HOST / DB_NAME / DB_PASSWORD]
+
+    C --> D[Load Data from DB\nsrc/yolo/db_dataset.py]
+    D --> D1[Scan temp/doc_id/images/\nfor rendered PNGs]
+    D --> D2[Batch load field_results\nfrom database - batch 500]
+
+    D1 --> E[Create DBYOLODataset]
+    D2 --> E
+
+    E --> F[Split Train/Val/Test\n80% / 10% / 10%\nDocument-level, seed=42]
+
+    F --> G[Export to YOLO Format]
+    G --> G1[Copy images to\ntrain/val/test dirs]
+    G --> G2[Generate .txt labels\nclass x_center y_center w h]
+    G --> G3[Generate dataset.yaml\n+ classes.txt]
+    G --> G4[Coordinate Conversion\nPDF points 72DPI -> render DPI\nNormalize to 0-1]
+
+    G1 --> H{--export-only?}
+    G2 --> H
+    G3 --> H
+    G4 --> H
+
+    H -- Yes --> Z[Done - Dataset exported]
+    H -- No --> I[Load YOLO Model]
+
+    I --> I1{--resume?}
+    I1 -- Yes --> I2[Load last.pt checkpoint]
+    I1 -- No --> I3[Load pretrained model\ne.g. yolo11n.pt]
+
+    I2 --> J[Configure Training]
+    I3 --> J
+
+    J --> J1[Conservative Augmentation\nrotation=5 deg, translate=5%\nno flip, no mosaic, no mixup]
+    J --> J2[imgsz=1280, pretrained=True]
+
+    J1 --> K[model.train\nUltralytics Training Loop]
+    J2 --> K
+
+    K --> L[Training Outputs\nruns/train/name/]
+    L --> L1[weights/best.pt\nweights/last.pt]
+    L --> L2[results.csv + results.png\nTraining curves]
+    L --> L3[PR curves, F1 curves\nConfusion matrix]
+
+    L1 --> M[Test Set Validation\nmodel.val split=test]
+    M --> N[Report Metrics\nmAP@0.5 = 93.5%\nmAP@0.5-0.95]
+
+    N --> O[Close DB Connection]
+
+    style A fill:#4a90d9,color:#fff
+    style K fill:#e67e22,color:#fff
+    style N fill:#27ae60,color:#fff
+    style Z fill:#95a5a6,color:#fff