IDDA V2

We are proud to announce the second release of our IDDA dataset designed for both Semantic Segmentation and Visual Place Recognition task. IDDA V2 is a new synthetic dataset that includes both GPS/IMU information and pixel-wise semantic annotations for 25 semantic classes (with 17 of them in common to the Cityscapes standard). This second release contains more than 40000 images (10091 per scenario) captured across two different urban maps (Town03 and Town10 from CARLA notation) and in two weather conditions, Clear Noon (CN) and Hard Rain Sunset (HRS). To collect the data we equipped the ego-vehicle with four cameras (front, rear, left and right view). We split the front-view and rear-view frames captured in the Town10 in a gallery and a query set.

For better understanding read our paper, get the code, download IDDA V2 and start working with it!