Lesson 02: Installation & First Labels

Three installation paths. Pick the one that matches your setup.

Method 1: pip (Recommended)

Requires Python 3.10 or later. Python 3.12 is recommended.

# Create an isolated environment
conda create -n anylabeling python=3.12
conda activate anylabeling

# CPU version
pip install anylabeling

# Launch
anylabeling

For GPU-accelerated inference (NVIDIA GPUs with CUDA):

pip install anylabeling-gpu
anylabeling

The GPU version uses ONNX Runtime with CUDA, which makes SAM and YOLO inference significantly faster — especially for large images or ViT-H models.

macOS note: If you hit PyQt issues on macOS, install PyQt6 via conda first: conda install pyqt6 before pip install anylabeling.

Method 2: Pre-Built Binaries

Download the latest release from GitHub Releases. Releases tagged with -GPU include CUDA support.

• Windows: Download the .exe, double-click to run.
• macOS: Download the folder-mode build (not the single-file .app).
• Linux: Download the AppImage or extract the tarball.

No Python installation required. This is the fastest path if you just want to label images without touching a terminal.

Method 3: From Source

For development or bleeding-edge features:

git clone https://github.com/vietanhdev/anylabeling
cd anylabeling
pip install -e .
python anylabeling/app.py

The Interface at a Glance

When you launch AnyLabeling, you see four main areas:

Tool Bar (left) — annotation tools (rectangle, polygon, point, etc.) and the brain icon for AI auto-labeling.

Canvas (center) — the image you are annotating. Zoom with Ctrl + scroll wheel. Pan by holding Ctrl and dragging.

Label List (top right) — all labels/annotations on the current image. Click to select, double-click to rename.

File List (bottom right) — all images in the current directory. Click to navigate. Current image is highlighted.

Loading Images

Three ways to load images:

Ctrl + U is the one you will use most. It opens every image in a directory and lists them in the File List panel.

Your First Annotation

Let's draw a bounding box:

1. Open a directory with images (Ctrl + U).
2. Select Rectangle from the toolbar, or press R.
3. Click and drag on the canvas to draw a rectangle around an object.
4. A dialog appears — type the label name (e.g., car) and press Enter.
5. The annotation appears in the Label List. The label file is auto-saved.

That is it. The annotation is saved as a .json file alongside your image, in LabelMe-compatible format. Every time you draw, edit, or delete a shape, the file updates automatically.

Essential Keyboard Shortcuts

These shortcuts will save you hours over the course of a real annotation session:

Tip: Enable View > Keep Previous Zoom Ratio to maintain your zoom level when switching between images. This is essential when annotating small objects across a dataset.

The Label File Format

AnyLabeling saves annotations in a JSON format compatible with Labelme:

{
  "version": "0.3.3",
  "flags": {},
  "shapes": [
    {
      "label": "car",
      "points": [
        [100, 150],
        [400, 350]
      ],
      "group_id": null,
      "shape_type": "rectangle",
      "flags": {}
    }
  ],
  "imagePath": "image_001.jpg",
  "imageHeight": 720,
  "imageWidth": 1280
}

Each image gets its own .json file. This is the internal format — you will export to COCO, YOLO, or VOC for training in Lesson 07.

Adjusting Display Settings

Two settings that matter for annotation quality:

Brightness and contrast — View > Brightness Control and View > Contrast Control. Useful for dark or overexposed images. Enable Keep Current Brightness and Keep Current Contrast to apply adjustments across all images in the session.

Cross-line cursor — helps with precise alignment. Toggle it from the View menu.

Verify Your Installation

Run through this checklist:

1. AnyLabeling launches without errors.
2. You can open a directory of images with Ctrl + U.
3. You can draw a rectangle and assign a label.
4. The .json file appears next to your image.
5. You can navigate between images with A and D.

If everything works, you are ready. In the next lesson, we go deep on every annotation type AnyLabeling supports — polygons, points, lines, circles, and rotated bounding boxes — with the shortcuts and techniques that make manual annotation fast.