# Running YOLO Scripts Manually

This guide explains how to extract, configure, and manually execute the Python runner scripts (`predict.py` and `train.py`) provided by **DiGi.YOLO**.

---

## 🛠️ Prerequisites & Installation

Before running the scripts, ensure Python (3.10+) is installed and install the required dependencies:

```bash
pip install -r requirements.txt
```
*(Dependencies: `ultralytics`, `torch`)*

---

## 📂 Exporting Scripts via C#

Call `Modify.WriteScripts(targetDirectory)` to export all runner scripts (`train.py`, `predict.py`, `utils.py`, `requirements.txt`) and configuration (`conf.yaml`) to your working directory:

```csharp
using DiGi.YOLO;

string targetDirectory = @"C:\Path\To\WorkingDir";
bool success = Modify.WriteScripts(targetDirectory);
```

---

## 🔍 Running `predict.py` (Inference)

`predict.py` uses `argparse` to allow explicit control over input image sources, model weights, confidence thresholds, and output bounding box result files (`.bbrf`).

### Command-Line Arguments

| Flag | Default Value | Description |
|---|---|---|
| `--model` | `YOLO/models/model.pt` | Path to trained YOLO model weights (`.pt`). Defaults to best checkpoint in `runs/detect/train*/weights/best.pt` if default does not exist. |
| `--source` | `YOLO/input` | Path to single input image file or directory containing `.jpg`, `.jpeg`, or `.png` images. |
| `--conf` | `0.1` | Confidence threshold float value for detection filtering. |
| `--output` | `YOLO/output/results.bbrf` | Destination path for output bounding box result file (`.bbrf`). Written in `"w"` mode to prevent duplicate rows. |

### Usage Examples

#### Default Run
```bash
python predict.py
```

#### Custom Arguments
```bash
python predict.py --model YOLO/models/model.pt --source "C:/Data/Images" --conf 0.25 --output "C:/Data/Output/results.bbrf"
```

---

## 🏋️ Running `train.py` (Training)

`train.py` trains or fine-tunes a YOLO model using dataset configuration defined in `conf.yaml`.

### Directory Structure Requirements

Keep Python scripts and configuration files in the root execution directory, with dataset files structured under a common `training/` folder:

```text
.
├── conf.yaml
├── predict.py
├── requirements.txt
├── train.py
├── utils.py
└── training/
    ├── images/
    │   ├── train/
    │   └── val/
    └── labels/
        ├── train/
        └── val/
```

### Dataset Configuration (`conf.yaml`)

```yaml
path: training
train: images/train
val: images/train
names:
  0: building
```

### Execution

Run training with default hyperparameters (150 epochs, `imgsz=640`):

```bash
python train.py
```

The training process outputs model checkpoints to `runs/detect/train*/weights/best.pt`.