CARLA Simulator Driving Scenes for CNN Training

A collection of image data generated using the CARLA autonomous driving simulator for training convolutional neural networks. The dataset is sourced from Kaggle, a platform for data science competitions and datasets. Specific details on the number of images, time of creation, and original author are not provided in the input metadata.

Use Cases

• Train a CNN for vehicle detection using bounding box labels on simulated urban scenes.
• Develop semantic segmentation models for road, sidewalk, and building classification from pixel-level annotations.
• Validate perception algorithms on synthetic data featuring varied lighting and weather conditions like rain or fog.
• Perform domain adaptation research by comparing model performance on CARLA-simulated images versus real-world driving datasets.

Strengths

• Data generated from the CARLA simulator, which provides controlled, photorealistic driving scenarios.
• Likely includes diverse environmental conditions (e.g., time of day, weather) that are expensive to capture in the real world.

Limitations

• Unknown total number of images or scenes, making it difficult to assess scale.
• Simulated data may contain artifacts and lack the full complexity and noise of real-world sensor data, leading to sim-to-real gap issues.

Provenance

Source

CARLA autonomous driving simulator.

Collection Method

Synthetic data generation from a simulation environment.

Geography

Simulated urban environments.