# YOLOv9 dataset augmentation

Augment a YOLOv9-format dataset by creating new image and label files for horizontal flip, vertical flip, +10% hue, +30% contrast, and grayscale. Labels are updated correctly for flips; other augmentations copy labels unchanged.

## Dataset layout

Expected structure:

```
dataset/
├── images/          # .jpg or .png
│   ├── img1.jpg
│   └── img2.jpg
└── labels/          # .txt, one per image, same base name
    ├── img1.txt
    └── img2.txt
```

YOLO label format: one line per object: `class_id x_center y_center width height` (normalized 0–1).

If `images/` and `labels/` are not present, the script treats the given directory as containing both images and labels (flat layout).

## Setup

```bash
pip install -r requirements.txt
```

## Usage

Augment in place (new files appear next to originals in `images/` and `labels/`):

```bash
python augment_yolov9_dataset.py --dataset-dir ./dataset/train
```

Write augmented files to a separate directory (creates `train_aug/images/` and `train_aug/labels/`):

```bash
python augment_yolov9_dataset.py --dataset-dir ./dataset/train --output-dir ./dataset/train_aug
```

Other options:

- `--image-ext .png` — look for `.png` instead of `.jpg`
- `--suffixes hflip vflip` — run only horizontal and vertical flip (choices: `hflip`, `vflip`, `hue`, `contrast`, `gray`)
- `--dry-run` — print which files would be created without writing

## Output naming

For each image `img.jpg` with label `img.txt`, the script can create:

| Augmentation   | Image           | Label           |
|----------------|-----------------|-----------------|
| Horizontal flip| `img_hflip.jpg` | `img_hflip.txt` |
| Vertical flip  | `img_vflip.jpg` | `img_vflip.txt` |
| Hue +10%       | `img_hue.jpg`   | `img_hue.txt`   |
| Contrast +30%  | `img_contrast.jpg` | `img_contrast.txt` |
| Grayscale      | `img_gray.jpg`  | `img_gray.txt`  |

Add these paths to your YOLOv9 data YAML or file lists to use the augmented set.