Skip to main content

Image Dataset Format (Image Detection)

Image Detection​

Each image must have a corresponding label file. Expected folder structure:

datasets/
└── human_face/            # one folder per class
├── image1.jpg
├── image2.jpg
├── labels/            # YOLO .txt annotation files
│   ├── image1.txt
│   └── image2.txt
└── classes.txt        # class names for this folder

Label file format — one line per bounding box:

<class_id> <x_center> <y_center> <width> <height>
  • All coordinate values normalised between 0.0 and 1.0
  • class_id is zero-based, matching your classes.txt file
NOTE

If a label file is missing for an image, the tool treats that image as unannotated. Use the Dataset Health tab to find and manage unannotated images before training.