The dataset contains 11,913 frame pairs of urban driving footage with and without moving objects, synthetically generated with the CARLA simulator. All frames are available both as RGB images and semantic segmentations. RGB images are non-photorealistic being rendered by a game engine, while semantic segmentations are similar to a real-world segmentations. The dataset is designed to provide a supervision for semantic road inpainting tasks.

Samples

License

  • Unknown

Modalities

Languages

Tasks