Notable features

Type of data:

  • RGB
  • Pixel-wise Semantic annotation
  • Depth map
  • Semantic RGB Cityscapes-palette conversion
  • Depth Gray-scale conversion
  • Logarithmic Depth Gray-scale conversion

Dataset Dimensions:

  • 1,006,800 images for each data type
  • 1920x1080 pixels
  • FoV of 90°
  • 1TB of memory occupation

Data Diversity:

  • 6 cities + 1 bucolic country
  • 3 weather and illumination conditions:
    • Clear Noon
    • Clear Sunset
    • Hard Rain Noon
  • 5 different camera heights (from lower to higher position)
  • 5 viewpoints with and without hood (if visible, hood different in shape and color):
    • Audi TT
    • Ford Mustang
    • Jeep Wrangler Rubicon
    • Volkswagen T2
    • Bus
  • 105 different scenarios with around 16k training/testing images for each
  • Always different distributions of vehicles and pedestrians in the scenes

Semantic Segmentation

  • 24 semantic classes
  • 2087.70 x 109 annotated pixels

    Classes Definition:

    Class Name Description
    0* None all the remaining objects not labeled
    1 Building skyscraper, house, bust stop, garage, bridge,fountains and other types of constructions
    2 Fence barrier, railing or other upright structure
    3* Other all static objects not classified, but not None
    4 Pedestrian human person (male, female or kids). Not included what is carried by the person
    5 Pole vertical oriented piece of wood or metal
    6* Road line all the marks on the road, line painted on the pavements
    7 Road any kind of drivable road
    8 Sidewalk part of the ground designated only for pedestrian
    9 Vegetation tree, hedge, plants
    10 Vehicles cars and truck, visible hood
    11 Wall vertical brick or stone structure
    12 Traffic Sign the part of the sign containing the information, not the pole
    13 Traffic Light the traffic light box, without pole
    14* Guardrail crash barriers
    15* Dynamic movable objects like trash, bin, bag or wheelchair
    16 Bicycle cross bikes, leisures bike and road bikes
    17 Motorcycle four kind of motorbikes
    18 Rider rider of bike or motorcycle, not pedestrian
    19 Terrain background, grass, soil, sand or rocks
    20 Sky the open sky
    21* Railway rail truck or elevated rail line
    22* Ground horizontal ground-level structure
    23* Static clutter in the background not distinguishable, like mountains

    *This class is not considered during evaluation

RGB Image Example Semantic Image Example Depth Image Example
Click to see some sample taken from IDDA:RGB, Semantic, and Depth.