Skip to content

Latest commit

 

History

History
82 lines (74 loc) · 3.22 KB

synthia.md

File metadata and controls

82 lines (74 loc) · 3.22 KB
Raw

SYNTHIA [1] is a synthetic dataset of urban scenes, created from a virtual city. It has pixel-level annotations of 13 classes:

sky, building, road, sidewalk, fence,
vegetation, pole, car, sign, pedestrian,
cyclist, lane-marking, miscellaneous

It consists of several subsets (see the Downloads page in [1]):

  • SYNTHIA-Rand. Introduced in [1], it contains 13,407 images of size 960x720. Those images are not temporally consistent because this subset is not a video stream. Images are annotated in class level, but not instance level, with 11 classes (and a void):

    void, sky, building, road, sidewalk,
fence, vegetation, pole, car, sign,
pedestrian, cyclist

  • SYNTHIA-Rand-Cityscapes. Introduced in [1], it contains 9,400 images of size 720x1280. Images are also not temporally consistent. This subset provides instance level annotations of 23 classes:

    void, sky, building, road, sidewalk,
fence, vegetation, pole, car, traffic sign,
pedestrian, bicycle, motorcycle, parking-slot, road-work,
traffic light, terrain, rider, truck, bus,
train, wall, lanemarking

    This class set is compatible with the Cityscapes test set.

  • SYNTHIA-Seqs / SYNTHIA Video Sequences. Introduced in [1], it contains 7 videos obtained from 8 views/cameras with frame size 960x720. Each video is further divided into sub-sequences with same traffic situation but under different weather/illumination/season condition. This subset provides instance level pixel annotations of 13 classes:

    sky, building, road, sidewalk, fence,
vegetation, pole, car, sign, pedestrian,
cyclist, lane-marking, miscellaneous

  • SYNTHIA-SF. Introduced in [2], it contains 6 videos, each associating left & right image, instance level pixel annotations and depth. The class set is also Cityscapes compatiable:

    void, road, sidewalk, building, wall,
fence, pole, traffic light, traffic sign, vegetation,
terrain, sky, person, rider, car,
truck, bus, train, motorcycle, bicycle,
road lines, other, road works

  • SYNTHIA-AL. Introduced in [3], it contains 1 video with annotations of instance level, 2D bounding boxes, 3D bounding boxes and depth. The class set is:

    void, sky, building, road, sidewalk,
fence, vegetation, pole, car, traffic sign,
pedestrian, bycicle, lanemarking, traffic light

References

  1. (CVPR 2016) The SYNTHIA Dataset: A Large Collection of Synthetic Images for Semantic Segmentation of Urban Scenes - paper, project, paper with code
  2. (BMVC 2017) Slanted Stixels: Representing San Francisco's Steepest Streets - paper, blog
  3. (ICCV Workshop 2019) Temporal Coherence for Active Learning in Videos - paper