Dataset for Autonomous Cars — Annotated Images
More than 22,000 annotated images from real driving situations, ideal for training embedded vision models for autonomous vehicles.
22,200 annotated images, tags: car, truck, truck, pedestrian, cyclist, traffic light
CC0: Public Domain
Description
The dataset Self-Driving Cars includes over 22,000 images annotated with key categories such as car, truck, pedestrian, cyclist, and traffic light. These images were originally produced by Udacity and are particularly suited to object detection tasks in autonomous driving contexts.
What is this dataset for?
- Training real-time object detection models for autonomous cars
- Test the accuracy of segmentation or visual classification models
- Simulate driving environments in AI frameworks
Can it be enriched or improved?
This dataset can be enriched by adding finer annotations (bounding boxes, pixel-level segmentation). It is also possible to integrate weather, hourly or geographic data to make learning more complex. For optimization, the increase in data (fog, blur, night) makes it possible to simulate various driving conditions.
🔎 In summary
🧠 Recommended for
- AI mobility engineers
- Embedded software developers
- Autonomous perception researchers
🔧 Compatible tools
- Yolov5/v8
- Detectron2
- TensorFlow Object Detection API
💡 Tip
For more realistic results, combine this dataset with video sequences in real driving or simulators (CARLA).
Frequently Asked Questions
Does this dataset include bounding box coordinates?
No, only object classes are provided. However, it is possible to add manual annotations if necessary.
Can this dataset be used in a commercial context?
Yes, the CC0 license allows unrestricted reuse, even for commercial purposes.
Is this a good starting point for a Yolo detection model?
Yes, especially for learning basic categories; adding bounding boxes is recommended to make the most of it.
