Dataset Partition Description

Cloud-Robotics dataset has 7 groups of 30 classes. The partition of dataset is as follow:

Group Classes
flat road · sidewalk · ramp · runway
human person · rider
vehicle car · truck · bus · train · motorcycle · bicycle
construction building · wall · fence · stair · curb · flowerbed · door
object pole · traffic sign · traffic light · CCTV camera · Manhole · hydrant · belt · dustbin
nature vegetation · terrain
sky sky

Data Statistics

Features

Polygonal annotations

  • Dense semantic segmentation
  • Instance segmentation for vehicle and people

    • Complexity

  • 30 classes

    • Diversity

  • Daytime
  • Manually selected frames
  • Large number of dynamic objects
  • Varying scene layout
  • Varying background
  • Volume
  • 2600 annotated real-world images
  • Benchmark suite and evaluation server
  • Pixel-level semantic labeling

    • Data Format

    The format of json annotation is as follows: 

    "gtFine_polygons.json": {
        "imgHeight": <int>        -- Height of the image
        "imgWidth": <int>         -- Width of the image
        "objects": <list>         -- List of the objects
        "label": <str>            -- Name of the label
        "polygon": <list>         -- Boundary point coordinates
}

    Data Annotation

    Below are examples of high quality dense pixel annotations of 2500 real-world images. Overlayed colors encode semantic classes. Note that single instances of traffic participants are annotated individually.