I’m excited to announce Synthetic YOLO Dataset Generator, an editor extension for Unity that creates high-quality synthetic image datasets for computer vision. If you need thousands of labeled images for training YOLO object detection or segmentation models, this tool can save you tons of time – no coding required.
What it does: It allows you to generate thousands of randomized scenes and automatically produce annotated images (bounding boxes, polygon masks in YOLO format) inside the Unity Editor. You set up your 3D objects as prefabs, configure some parameters in the Inspector, and hit “Generate” – the tool handles the rest (spawning objects, randomizing lighting/rotations, applying post-effects, and capturing labeled screenshots).
Key Features:
-
YOLO-format Datasets: Exports images with corresponding TXT label files for object detection, plus JSON polygon annotations for segmentation (compatible with YOLOv8–YOLOv12 formats). It automatically creates train/val/test splits in structured folders.
-
Fully Customizable Scenes: Works with any Unity 3D models or prefabs. You can randomize object placement, rotation (with control over axes), camera angles, lighting conditions, and even apply post-processing effects like Depth of Field, Motion Blur, Color Grading, etc., to mimic real-world variability.
-
One-Click Generation: Everything is configured in the Unity Inspector – camera settings, resolution, output paths, how many images to generate, etc. – so you don’t have to write any code or external scripts. Just tweak the settings and generate.
-
Quick Visual Feedback: The tool creates a “Labeled” folder with sample images overlayed with their annotations, so you can quickly verify that the bounding boxes and segmentations line up correctly (super useful for debugging your dataset setup!).
Use Cases: Perfect for Unity developers or researchers working on:
-
Autonomous vehicles or robotics – simulate environments to train object detection without risky field data collection.
-
Security cameras / surveillance AI – generate various scenario images with labeled intruders or objects of interest for model training.
-
AR projects or games with ML features – quickly prototype computer vision models by creating synthetic training data for your game objects.
-
Academia – if you need a custom vision dataset (e.g., for a research project), you can build it in Unity with diverse asset packs, instead of hand-labeling thousands of images.
The package includes a detailed PDF User Guide and we’ve prepared a YouTube video tutorial (link below) to help you get started quickly. No additional libraries needed aside from Unity (URP is required for versions < 2020.1). This has been tested on Unity 2021–2022 and works with both the Built-in and URP render pipelines.
You can find Synthetic YOLO Dataset Generator on the Unity Asset Store: Asset Store Link. The YouTube demo is available here: https://youtu.be/lB1KbAwrBJI. Feel free to check it out, and I’d love to hear any feedback or answer questions!
